The use of machine learning to predict the main factors that influence the continuous usage of mobile food delivery apps

Abstract

The popularity of mobile food delivery apps (MFDAs) and the online food delivery industry surged during the COVID-19 epidemic. Despite the explosive growth in the use of these apps, relatively limited research has been done to determine what affects their continuous use. This study predicts the continuous use of MFDAs and explores the variables that influence this utilization using a novel machine learning (ML) based approach. The machine learning models included four distinct constructs (i.e., features): perceived compatibility, convenience, online reviews, and delivery experience. These features were measured using a survey instrument. Eight different machine learning (ML) models, ranging from basic decision trees to neural networks, were deployed. All eight models achieved high prediction accuracy of above 93%, with the CatBoost model having the highest accuracy among them at 98%. Feature importance analysis revealed perceived compatibility to be the most important factor impacting the continuous usage of MFDAs followed by convenience, online reviews, and delivery experience respectively. The study’s findings have ramifications for MFDA marketing and design. Given the significance of perceived compatibility, MFDA marketing campaigns should have a strong emphasis on highlighting how well these apps fit with the users’ lifestyles.

Keywords

Continuous intention COVID-19 food delivery apps machine learning

1. Introduction

By the end of 2019, the COVID-19 virus had rapidly spread around the world (Zhao & Bacao, 2020). To stop the COVID-19 virus from spreading, governments around the world put in place strict measures, including self-isolation, social distance, masking, and avoiding direct or indirect social connections (Kumar & Shah, 2021). To contain the outbreak and lessen its effects on the general public in the State of Kuwait, the study’s context, the government has put in place a number of progressive restrictive measures, including border closures, airport closures, the lockdown of non-essential businesses, home isolation, and ultimately a partial curfew for the entire nation (Rabaa’i, in press a). Because of this, the COVID-19 outbreak has had a significant negative impact on the conventional catering industry, which was predicted to have lost over 240 billion US dollars by the year 2020 (Rabaa’i, in press b). In reaction to the destruction COVID-19 has done on the catering industry, both globally and in Kuwait, many conventional catering businesses have shifted from traditional in-store service to online-to-offline (O2O) food delivery services. This has been done in order to meet customer expectations and stay in business during the crisis (Zhao & Bacao, 2020). Using mobile food delivery apps (MFDAs), “customers can browse a list of subscribed restaurants, their menus, and ratings, confirm orders via online payment, and monitor order statuses without having to speak to restaurants in person or on the phone” (Kaur et al., 2021). MFDAs are characterized as service-based mobile apps that let users place online food orders and have those orders delivered to their homes (Alalwan, 2020).

MFDAs are one of the mobile app categories that are expanding the fastest, according to Statistia (2020). The report projects that the global online food delivery market will generate 107.4 billion dollars in revenue in 2019 and 182.3 billion dollars in revenue by 2024. (Statistia, 2020). From 2019 to 2024, the MFDAs sector is projected to expand at a rate of 11.4 percent annually (IMARC, 2020). However, in order to continue this steady expansion, the industry must sustain and expand its user base. Information systems (IS) researchers have examined early usage and the variable that might affect the initial acceptance of MFDAs using a variety of theoretical models and frameworks, but they have not completely analyzed and accounted for changes in use patterns in the future (e.g., Elvandari et al., 2018; Gunden et al., 2020; Mehrolia et al., 2021; Roh & Park, 2019; Yeo et al., 2017). A detailed investigation into how usage behaviors evolve in the future is required since the long-term effectiveness and durability of a technology, like MFDAs, are characterized by sustained future use rather than initial adoption or usage (Bhattacherjee, 2001). Studies focusing on the continuous use of MFDAs are quite rare (e.g., Alalwan, 2020; Kaur et al., 2021). Only few studies have looked at the variables that influence the ongoing usage of MFDAs (Al Amin et al., 2021; Alalwan, 2020; Cho et al., 2019; Kumar & Shah, 2021). The majority of these research studies (Roh & Park, 2019), which primarily concentrate on the technological advantages or drawbacks of these apps, show conflicting results, and neglect to take into account the factors that constitute the particular context of MFDAs continuous usage (Rabaa’i, in press b).

Structural equation modeling (SEM) has traditionally been utilized to study the continuous use of MFDAs. The objective of this study is to use novel, contemporary and cutting-edge machine learning (ML) techniques to predict users’ continuous intention to use MFDAs, thus contributing to the existing body of literature. The ML models incorporate pertinent variables as predictors that indicate the specific context of MFDAs continuous usage, such as convenience, perceived compatibility, delivery experience, and online reviews. To predict the continuous usage of MFDAs and analyze these factors, we implemented several ML techniques, including Decision Tree, Random Forest, Support Vector Machines, LightGBM, CatBoost, XGBoost, Bagging Classifier, and Artificial Neural Networks (ANN).

The reminder of the paper is structured as follows: The theoretical underpinnings, the suggested research model, and the constructs employed in this study are presented in Section 2. The research methodology is described in Section 3 followed by the data analysis and findings in Section 4. Discussions and implications of the results are addressed in Section 5. The limitations and further research directions are discussed in the conclusion.

2. Theoretical foundation and the constructs used in this study

Information Systems researchers have employed different theoretical models and frameworks, such as the technology acceptance model (TAM) (Davis et al., 1989), Unified Theory of Acceptance and Use of Technology (UTAUT) model (Venkatesh et al., 2003, 2012), and the theory of planned behavior (TPB) (Ajzen, 1991) to examine MFDAs continuous usage – Detailed literature review related to MFDAs can be found in Rabaa’i (in press b). For example, Okumus and Bilgihan (2014) investigated what influences users’ decisions to use their smartphones to place food orders and discovered that perceived usefulness, ease of use, perceived enjoyment, self-efficacy, and social norms are the primary deciding factors. According to Lee et al. (2017), consumer perceptions toward MFDAs are predicted by perceived utility and simplicity of use. Also, Yeo et al. (2017) argued that convenience motivation, post-usage usefulness, hedonic incentive, price and time-saving orientation, previous purchase experience, and attitudes all affect users’ intention to use MFDAs. Elvandari et al. (2018) suggested that order conformity, cleanliness of food box, excellent condition of the received ordered food, politeness and friendliness of delivery staff, and affordable delivery costs influence customers’ behavioral intention toward using MFDAs. Additionally, Ray et al. (2019) reported that the key gratifications associated with MFDAs are ease of use, delivery experience, customer experience, and convenience. Their study discovered that consumers’ decisions to buy food via MFDAs are significantly influenced by the user experience, usability, and listings and searches for restaurants. Roh and Park (2019) reaffirmed prior research findings and confirmed the effect of usefulness and compatibility on the intention to use MFDAs. Troise et al. (2020) integrated TAM and TPB and found that personal attitude, subjective norms, trustworthiness, and the perception of risks were positively correlated with the behavioral intentions to use MFDAs. Similarly, Mehrolia et al. (2021) argued that customers with a greater perceived threat during COVID-19, less product participation, a lower perceived advantage of food delivery, and less online food orders are allegedly less inclined to use MFDAs to place food orders.

In contrast to studies investigating initial use and adoption of MFDAs, fewer studies have investigated the variables that impact the continuous usage of MFDAs. Lee et al. (2019) applied the UTAUT model and reported that information quality, habit, and social influence are crucial in continued usage intentions of MFDAs. Similarly, Cho et al. (2019) found that convenience, trustworthiness, design, various food choices are vital variables that affect users’ perceived value, attitudes and continuous usage of MFDAs Alalwan (2020) emphasized the impact of e-satisfaction and continuous intention to reuse MFDAs on the role of online reviews and ratings, online tracking, performance expectancy, hedonic incentive, and pricing value. Further, Zhao and Bacao (2020) proposed that continuous use of MFDAs is positively influenced by satisfaction, perceived task-technology fit, trust, performance expectancy, social influence, and confirmation. In the same vein, Al Amin et al. (2021) investigated the effects of behavioral and continuous intention to use MFDAs on social isolation, food safety, delivery hygiene, subjective norms, dining attitudes, and behavioral control. Their conclusions showed that behavioral and continuous intention to use MFDAs were associated with delivery hygiene, subjective standards, attitudes, and behavioral control. Finally, Kumar and Shah (2021) looked into how app aesthetics affect the feelings that MFDAs will have on their intentions to continue using the app. The findings showed that app aesthetics cause customers to feel pleasure, arousal, and domination, with pleasure being the most crucial predictor of continuous usage.

The above literature review clearly shows that there is a research gap in examining users’ continuous usage of MFDAs (Alalwan, 2020). Except for a small number of studies, the vast bulk of MFDAs research studies investigated behavioral usage intention rather than continuous usage. However, early acceptance or usage of a technology has not always led to continued use, as shown by the fact that some people initially embraced technology but later ceased utilizing it (e.g., Rabaa’i et al., 2021; Rabaa’i & Abu ALMaati, 2021). It has been argued that users’ intention to use a technology just demonstrates their favorable opinion about it and their corresponding desire to explore it (Huang, 2019), but it does not mean that they will utilize it indefinitely (Wang et al., 2019). Consumers will utilize a technology when it satisfies their requirements and expectations (Huang, 2019; Rabaa’i et al., 2022), and they will stop using it if and when it does not (Bhattacherjee, 2001).

In order to evaluate the variables affecting the continuous use of MFDAs, this study suggests a novel model. The explanation of each variable in the model and the theoretical justification for its inclusion are discussed below.

2.1 Convenience (CON)

Convenience (CON) has been a prevalent concept in traditional as well as online shopping (Lai & Liew, 2021) and an important factor in predicting users’ behavioral intentions (Rabaa’i, in press a, in press c). Kim et al. (2010, p. 314) argued that “convenience is related to the elements generating time and place utility for users”. That is, a product or service is considered convenient if it saves users time and alleviates their cognitive, emotional, and physical difficulties (Lai & Liew, 2021). Yeo et al. (2017) found a positive relationship between behavioral intentions to use online food delivery and perceived convenience. The authors argued that “consumer perceptions become positive when they are able to avoid dealing with the physical burden of traveling” (Yeo et al., 2017, p. 157). MFDAs provide clients with the convenience of comparing prices across restaurants, eliminating wait times, and avoiding traffic-related issues (Ray et al., 2019). Past studies (e.g. Cho et al., 2019; Rabaa’i, in press b; Ray et al., 2019; Roh & Park, 2019) verified that convenience has a very positive effect on users’ behavioral intentions while using food delivery services. As such, it is reasonable to argue that users will regard such technology as useful and convenient, hence motivating their continuous use intention, if MFDAs enable consumers to order food more effectively and efficiently by lowering wait times, avoiding traffic, and facilitating comparative shopping.

2.2 Perceived compatibility (PC)

According to Rogers (2003), how well a technology matches a person’s working style, lifestyle, values, and needs is known as perceived compatibility (PC). A perception of high compatibility will result in rapid adoption of a technology in general (Pham & Ho, 2015), and MFDAs in particular. In mobile payments context, for example, Chen (2008, p. 39) defined PC as “the extent to which m-payment is consistent with the prospective user’s lifestyle and the way he or she likes to shop”. According to Rabaa’i (in press b), PC was strongly correlated with the continuous intentions to use MFDAs. Previous studies (e.g., Marinković et al., 2020; Roh & Park, 2019) have confirmed significant relationship between PC and behavioral intentions towards different technologies. Therefore, the current study asserts that if MFDAs users believe that such apps are compatible with their lives and the way they like to shop, they will continue to use them in the future.

2.3 Online reviews (OR)

Positive or negative, OR are a “sort of electronic word-of-mouth (eWOM) communication” (Qahri-Saremi & Montazemi, 2022). OR are defined as “peer-generated evaluations about products or services posted on retailer or third-party websites” (Hong et al., 2018, p. 1). These OR can be functional in reporting on the service efficiency of a restaurant or emotional in that they express sentiments about the restaurant’s service experience (Aureliano-Silva et al., 2021, p. 1758). Customers can use MFDAs to generate and share OR and feedback about the restaurants from which they ordered food, which can then help other customers in deciding where to order from using their MFDAs (Alalwan, 2020). When purchasing a product or comparing alternatives on the same online app, customers consider these reviews to be very important (Alalwan, 2020; Aureliano-Silva et al., 2021). Rabaa’i (in press b) argued that OR are directly linked to product sales, forming customers’ opinions, and assisting customers in making decisions. Prior studies (e.g., Alalwan, 2020; Aureliano-Silva et al., 2021; Hong et al., 2018) revealed a significant positive association between OR and behavioral intentions in various contexts. Hence, this study suggests that the reliability, relevance, and suitability of OR published on MFDAs will encourage future use of these platforms.

2.4 Delivery experience (DE)

MFDAs enable customers to track deliveries in real time, ascertain anticipated arrival times, locate the delivery destination on a map, take advantage of free delivery with some items (Ray et al., 2019, p. 225) and refrain from calling restaurants to find out the status of orders (Alalwan, 2020). When food is ordered through MFDAs, DE refers to the delivery experience (Rabaa’i, in press c). DE was crucial in motivating people to conduct online shopping (Kim et al., 2012). Previous studies (e.g., Alalwan, 2020; Maimaiti et al., 2018; Yeo et al., 2017) confirmed that DE stimulates customers’ behavioral intentions in the MFDAs context. Consequently, we argue that DE will induce users to continue using MFDAs in the future.

3. Methodology

A survey-based instrument was deployed to gather the required data. This section describes the measurement items, the study sample, and the data collection methodology.

3.1 Measurement instrument

Convenience, perceived compatibility, online reviews, delivery experience, and continuous intentions were the five study constructs that the survey instrument explored. The survey instrument had 17 items in all. Items for the questionnaire were taken from pertinent prior research studies, with a few phrasing adjustments made to fit the MFDAs context. The four delivery experience (DE) and three convenience (CON) measurement items were taken from Ray et al. (2019). For measuring online reviews (OR), all items were adapted from Alalwan (2020). Last but not least, all items for evaluating continuous intentions (CI) were taken from Bhattacherjee (2001). Each measurement item was assessed on a seven-point Likert scale, with (1) being “strongly disagree” and (7) being “strongly agree”.

3.2 Data collection and sample

Data for this study was collected during the COVID-19 epidemic in the State of Kuwait. Kuwait was selected as the context of this study due to the following reasons. First, MFDAs are a sector with great potential in Kuwait in comparison to other neighboring Gulf Council Countries (GCC) (Global Finance, 2020) Second, during the pandemic, Kuwaiti population used MFDAs more frequently due to a variety of factors, including the risk of contracting the disease, paranoia over the situation, long lineups at the cash register, and restricted access to actual stores. Third, as of September 2021, there are more than 4,000 restaurants in Kuwait, excluding multiple branches and cafes (Zawya, 2021), and approximately 3,600 of which already joined Talabat (Talabat, 2021) the dominant MFDA in Kuwait.

Students, alumni, instructors, and staff at a private American institution received an email inviting them to take part in the study along with a link to the online survey. They were also encouraged to invite their friends and family. Participants from outside the academic environment were also invited to participate in this study. The first researcher’s connections, who resided in Kuwait at that time, were sent the online questionnaire via several social media platform, in an effort to reach as many responders as possible. Due to government restrictions and the imposed curfew, paper-based data collection was not feasible during the outbreak. The validity of this study was ensured by the fact that every participant was an actual user of MFDAs. The screening question: “Do you use MFDAs when ordering food?” was employed to minimize the potential biases in the responses from respondents who do not use MFDAs. Respondents who provided a “no” response were excluded from this study. Three hundred and eighty-seven questionnaires were collected in total. Seventy-six were excluded through the screening question, leaving 311 valid survey responses.

The analysis of descriptive statistics was done using SPSS 23. 52 percent of respondents in the current study were females, and the majority were married (62 percent). Bachelor’s degrees were held by 52% of respondents. Only 7% of respondents were between the ages of 17 and 20 compared to 40% of respondents overall who were between the ages of 21 and 30. Students and working professionals made up the sample, with about 18 percent of students and 68 percent of working professionals. Finally, (80%) of the participants in the current study had used MFDAs for more than a year on average. Table 1 displays the descriptive statistics of the respondents.

Table 1
Demographic characteristics of respondents

Demographic variables		Frequency	Percentage
Gender	Male	159	48%
	Female	172	52%
	Total	331	100%
Age	17–20 years	24	7%
	21–30 years	133	40%
	31–39 years	98	30%
	$>$ 40 years	76	23%
	Total	331	100%
Marital status	Married	204	62%
	Single	127	38%
	Total	331	100%
Education level	High school	9	3%
	University student	52	16%
	Bachelor	149	45%
	Postgraduate	67	20%
	Others	54	16%
	Total	331	100%
Employment	Student	61	18%
	Working professionals	225	68%
	Unemployed	45	14%
	Total	331	100%
FDAs experience	$<$ 1 year	67	20%
	1–2 years	46	14%
	2–3 years	124	38%
	$>$ 3 years	94	28%
	Total	331	100%

4. Data analysis and results

This section discusses the reliability and validity of the instrument scale and the performance of the machine learning models used to predict the continuance intensions to use MFDAs.

4.1 Reliability and validity of the measurement scale

SmartPLS 3.2.9 software (Ringle et al., 2015) was utilized to assess the internal reliability, convergent as well as discriminant validity of the measurement scale. Internal reliability was assessed using Cronbach’s alpha (CA) and composite reliability (CR) (Hair et al., 2017). Table 2 shows that the CA and CR values for all constructs are above the recommended threshold of 0.7 and 0.85 respectively (Hair et al., 2019; Henseler et al., 2016). Factor loadings (FL) and the average variance extracted (AVE) were used to evaluate the convergent validity. Hair et al. (2017) suggested that FL and AVE values of each construct must be higher than 0.7 and 0.5, respectively. As shown in Table 2, FL ranging from 0.750 to 0.967, surpassed the recommended 0.7 threshold. According to Hair et al’s. (2017) rule of thumb, the AVE of all the constructs, in this study, which ranged from 0.727 to 0.916, explained more than 50% of the variance of their indicators (Henseler et al., 2009).

Table 2
Items loading, Cronbach’s alpha, composite reliability, and AVE

Items	Loading	Cronbach’s alpha	Composite reliability	AVE
Convenience (CON)		0.954	0.970	0.916
CON1	0.967
CON2	0.956
CON3	0.936
Perceived compatibility (PC)		0.875	0.923	0.800
PC1	0.889
PC2	0.904
PC3	0.891
Online review (OR)		0.874	0.914	0.727
OR1	0.826
OR2	0.824
OR3	0.901
OR4	0.858
Delivery experience (DE)		0.917	0.942	0.801
DE1	0.857
DE2	0.885
DE3	0.930
DE4	0.908
Continuous intention (CI)		0.841	0.906	0.764
CI1	0.750
CI2	0.929
CI3	0.931

Discriminant validity is “the extent to which a construct is empirically distinct from other constructs in the path model” (Sarstedt et al., 2014, p. 108), and can be evaluated by the Heterotrait-Monotrait (HTMT) criterion as suggested by Henseler et al. (2015). HTMT refers to “the mean value of the item correlations across constructs relative to the (geometric) mean of the average correlations for the items measuring the same construct” (Hair et al., 2019, p. 9). The results in Table 3 showed that all the HTMT values were lower than the recommended threshold of 0.90, thus confirming the discriminant validity (e.g. Alam et al., 2020; Hair et al., 2017; Rabaa’i et al., 2018, 2021; Rabaa’i & Zhu, 2021).

Table 3

Hetrotrait-monotrait ratio (HTMT) test

	CI	CON	DE	OR
CON	0.658
DE	0.807	0.559
OR	0.834	0.671	0.701
PC	0.739	0.678	0.633	0.805

4.2 Machine learning models

In predicting the continuous usage of MFDAs and to evaluate the significance of variables, we deployed a number of machine learning models. Machine learning models have been widely employed in predictive analysis in the field of information systems, particularly for binary classification issues (Akour et al., 2021, Uddin et al., 2019; Henrique et al., 2019). We examined eight popular machine learning models in this study to perform predictive analysis; they are briefly mentioned below.

Decision Tree algorithm is “a non-parametric supervised learning technique for classification” that learns from data-driven decision-making to predict the value of the target variable. Although it is frequently employed in decision analysis to openly and visually describe decisions and decision-making, it is also a widely utilized machine learning model for conducting predictive analysis (Alajali et al., 2018; Hasan et al., 2018). The Random Forest algorithm, an ensemble model, which extends decision trees and combines decorrelated trees with bootstrap aggregation (bagging), is frequently used in predictive analysis across a variety of research contexts (Iwendi et al., 2020; Speiser et al., 2019). Support Vector Machines (SVMs) are a group of supervised learning techniques used for outlier detection, regression, and classification. Each data item is represented as a point in n-dimensional space using n data features in the SVM algorithm. The process of categorizing involves identifying the hyper-plane that best separates the two classes. Frequently, SVMs models are used in binary classification problems (Cervantes et al., 2020; Sahu et al., 2015).

XGBoost model is a new technique offered by Chen and Guestrin (2016). Through additive training techniques, this model is designed to create a “strong” learner by merging all of the predictions made by a group of “weak” learners. The parallel calculations used in this model increase computational speed. Light Gradient Boosting Machine (LightGBM) model is a recently developed technique by Ke et al. (2017) that use exclusive feature bundling (EFB) and gradient-based one-side sampling (GOSS) to decrease the quantity of features and data instances. CatBoost algorithm is also a novel model that handles categorical features. (Dorogush et al., 2018). It overcomes various issues of traditional gradient boosted decision trees (GBDT) models to better train the model. These three boosting methods have been widely used in classification and predictive analysis (Li & Mao 2014; Tanha et al., 2020).

The last two methods applied in this study are bagging classifier and Artificial neural networks. Bagging Classifier is an ensemble meta-estimator that applies basic classifiers to randomly selected portions of the initial dataset and then averages or votes the individual forecasts to create a final prediction. By adding randomization to the process of creating a black-box estimator and using the resulting ensemble, it can be utilized to lower its variance (Kabari & Onwuka, 2019). In the domains of AI, machine learning, and deep learning, Artificial Neural Networks (ANNs), which are constructed based on the function of the human brain, enable computer programs to recognize patterns and resolve common issues. Artificial neural networks (ANNs) are constructed from layers of nodes or neurons, starting with an input layer, one or more hidden layers, and an output layer. Each node, or artificial neuron, is interconnected with others and comes with a weight and threshold. Any node whose output rises above the specified threshold value is activated and starts sending information to the top layer of the network. Otherwise, no data is transmitted to the network’s next layer.

To analyze the performance of each machine learning model, we used several metrics to evaluate the prediction, including prediction accuracy, precision, recall and F-measure. Accuracy is an intuitive metric for assessing a model’s overall prediction ability. Precision is the proportion of positive class forecasts (intend to use food delivery app) that really fall into the positive class. Recall, commonly referred to as sensitivity, is used to quantify how many true positives are accurately identified. The harmonic mean of precision and recall is taken into account when calculating the F-measure. These four measures which are defined in Table 4, have been widely used in existing literature to assess how well ML models perform predictive analysis (Alshurideh et al., 2020; Dwivedi, 2018; Sarker et al., 2019).

Table 4
Prediction results using machine learning models

Model	Accuracy (%) $\frac{TP+TN}{TP+FN+TN+FP}$	Precision $\frac{TP}{TP+FP}$	Recall $\frac{TP}{TP+FN}$	F-measure $2\times\frac{\textit{Precision}\times\textit{Recall}}{\textit{Precision}+% \textit{Recall}}$
Decision tree	97.58	1.00	0.954	0.976
Random forest	97.58	1.00	0.954	0.976
SVM	93.54	1.00	0.877	0.934
XGBoost	96.77	1.00	0.938	0.968
LightGBM	95.97	1.00	0.923	0.960
Catboost	98.39	1.00	0.969	0.984
Bagging	93.54	1.00	0.877	0.934
ANN	97.58	1.00

We divided the data into training and testing sets, using 80% of the data as the training dataset to calibrate the model’s parameters and 20% of the data as the testing dataset to assess how well each prediction performed. We used the oversampling methodology to increase the number of the minority class data points to address the imbalance in the dataset because the majority of users intended to continue using MFDAs. Table 4 summarizes the evaluation metrics for all models. All eight machine learning models have accuracy score higher than 93%, with the highest accuracy 98.39% of Catboost model. The high accuracy indicates the effectiveness of using four factors (PC, CON, OR, DE) for predicting the continuous usage of mobile food delivery app. All models have precision value 100% showing that all positive cases are predicted correctly. In other words, machine learning algorithms can accurately predict if a customer intends to continue using MFDAs. In contrast, recall values for all models are less than one, with smallest value 0.877 on SVM model, and highest value 0.969 on Catboost model. The results indicate less accuracy while predicting people who do not intend to use mobile food delivery app. F-measures are all larger than 0.93 indicating an overall effective prediction as well.

Table 5

Feature importance analysis

Model	PC	CON	OR	DE
Decision tree	0.765	0.048	0.153	0.033
Random forest	0.502	0.315	0.126	0.058
XGBoost	0.702	0.106	0.165	0.027
LightGBM	0.950	1.000	0.540	0.350
Catboost	0.513	0.197	0.188	0.103
ANN	0.865	0.241	0.139	0.183

Note: Feature importance analysis is not applicable in Bagging and SVM algorithms since they are not node-based models.

In addition to the predictive analysis, we were also interested in the importance of the different features (i.e., variables) used in the prediction. All four features used in the predictive model, PC, CON, OR, DE, contribute to the prediction, but some contribute more than others. We used feature importance analysis to rank these features. To improve the performance of the analysis and obtain deeper understanding of the application domain, feature importance or feature selection is an important stage when developing machine learning models (Qian et al., 2022; Zien et al., 2009). The calculation of feature importance is based on Gini index which measures the impurity level of each node in classification trees. The normalized total reduction of the criterion caused by each feature is used to determine its relevance score, with a higher number indicating greater importance. Table 5 includes the feature importance score estimated for the different machine learning models. PC is the most important feature with highest score among all models (e.g., 0.765 in Decision Tree) except LightGBM which ranks PC as the second most important feature with score 0.950. CON is the second most important feature. Three models – Random Forest, Catboost, and ANN – rank CON as the second most important feature, and LightGBM ranks it as the most important feature with score 1. The third important feature is OR, where three models (Random Forest, Catboost, LightGBM) ranking it the third, and two models (Decision Tree and XGBoost) rank it as the second most important feature. The only exception is the ANN model which treats OR as the least important feature with score 0.139. Compared to the other three features, DE is the least important feature since all models except ANN rank it at the end. In summary, the results show that all four factors contribute to the accurate prediction of the continuous use intentions of MFDAs, but PC and CON contribute more than the other two factors.

5. Discussion and implications

The primary goal of this study was to investigate the main variables that can affect the continuous usage of MFDAs in the post-adoption phase. The study employed a novel approach using machine learning models to evaluate influential factors, such as CON, PC, DE and OR, that represent the unique context of MFDAs continuous usage. While several studies, as described in section 2 above, have investigated the continuous usage of MFDAs, this study is among the first to use a novel machine learning based approach to investigate the problem. ML models like decision trees and neural networks are non-parametric models that do not make strict assumptions, such as a linear assumption in the case of structural equation models, the traditional model of choice in such studies. Thus, arguably machine learning models can be more effective in examining the factors impacting the continued use of MFDAs as they are able to capture any non-linear relationships that may exist in the data. Thus, by offering a novel method of assessing the continuous use of technology, this study significantly contributes to the field. This study also empirically demonstrates the effectiveness of such a machine learning based approach by establishing that a very high accuracy (all eight ML models produced an accuracy above 93%, with the highest accuracy reaching 98%) can be achieved in predicting whether a customer will continue to use MFDAs. The fact that such a high level of prediction accuracy can be obtained with the ML models would be of interest to MFDA service providers, such as Talabat in Kuwait, and more generally to any other MFDA maker in the world who is concerned about maintaining and growing their user base. This study can inform similar research that could be undertaken by MFDA players in their local markets, in order to grow their business. The high prediction accuracy of the ML models would be of interest to restauranteurs as well, as they will be able to accurately predict what share of their business would come from MFDAs as opposed dining in.

The high accuracy obtained by the ML models also validates the choice and efficacy of the four factors (i.e., PC, CON, OR, DE) used to predict the continuous usage of MFDAs thus lending credence to our analysis of the importance of these factors. The feature importance analysis confirmed that all four factors used in the predictive model – PC, CON, OR DE – contribute to the prediction of the continuous usage of MFDAs with PC being the most important and OR being the least important. Findings of the feature importance analysis propose the following: First, it can be deduced that users who believe that utilizing such platforms is consistent with their lifestyles and the way they want to buy will be more inspired to keep using these apps. This highlights the significance of perceived compatibility in predicting the continuous usage of MFDAs. This finding has effects on how MFDAs are marketed. Advertising campaigns run through different media like television, newspapers, and social media platforms should emphasize how well these apps fit into users’ lifestyles. This finding could potentially have branding implications as well. MFDAs should consider incorporating their user’s lifestyle choices in their branding exercises. Second, the results imply that users who find MFDAs to be convenient in reducing the waiting time, avoiding traffic, and comparing food prices from different restaurants will be driven to use these platforms in the future. As such, service providers and restaurant owners should highlight such conveniences in their advertising and marketing campaigns. This finding has potential design implications for the app as well. App designers should give importance to simple, quick, intuitive, and user-friendly ways to comparison shop on the app. Third, online reviews were found to be an important feature in predicting the continuous usage of MFDAs. Consequently, the credibility, relevance, and usefulness of the online reviews, posted on MFDAs, will promote users’ continuous usage of these platforms in the future. Therefore, service providers and restaurant owners should motivate users to review their experience when ordering food through the apps. Service providers should also guarantee the authenticity of the reviews posted on their apps and provide creative ways of aggregating user reviews into ratings and scores that can help in the user’s decision making. Finally, the finding that although not as significant as the other factors, the capabilities of MFDAs to view anticipated delivery times, find delivery locations on a map, track deliveries in real-time, and provide free delivery for specific items still help predicting the continuous use of MFDAs suggests that service providers and restaurant owners should think about offering promotions like free or discounted delivery for some items. Also, design considerations include enhancing the functions of the apps that let consumers check expected delivery times and track deliveries in real time.

6. Conclusion, limitations and future research

This study introduces a new methodology to investigate the continuous utilization of technology by applying a machine learning-based strategy to study the continuous usage of MFDAs. Employing eight commonly used machine learning models, this study investigated the influence of delivery experience, online review, perceived compatibility, and convenience, on predicting the continuous use of MFDAs in Kuwait, where this topic has never been researched before. A high degree of prediction accuracy is achieved by all eight ML models thus validating this approach. Feature importance analysis suggests that perceived compatibility was the most important factor, followed by convenience, online reviews and delivery experience respectively. These findings have broad marketing and design implications for the MFDA maker and the service provider.

This work has some limitations, which suggest some intriguing directions for further investigation. First of all, unlike many other developing countries, Kuwait is a developed Middle Eastern country with a tech-savvy citizenry (Rabaa’i, in press a, b, and c). Future research should investigate the proposed model from a transnational and transcultural standpoint, taking additional factors into account. Second, the participants in this study were not chosen at random, therefore the findings cannot be easily generalized to a larger population. As a result, future studies could use a random sample technique. Thirdly, although this study looked into a number of variables that might have an impact on the continuous use of MFDAs, it did not cover all of them. Future research should take into account additional variables such quality features, food safety and hygiene, and perceived mobility in order to widen the scope of the current study. Finally, the sample size of this study may restrict how broadly the results may be applied.

References

Ajzen

(1991). The theory of planned behavior. Organizational Behavior and Human Decision Processes, 50(2), 179-211.

Akour

Alshurideh

Al Kurdi

Al Ali

, & Salloum

(2021). Using machine learning algorithms to predict people’s intention to use mobile learning platforms during the COVID-19 pandemic: Machine learning approach. JMIR Medical Education, 7(1), e24032.

Al Amin

Arefin

M.S.

Alam

M.R.

Ahammad

, & Hoque

M.R.

(2021). Using mobile food delivery applications during COVID-19 pandemic: An extended model of planned behavior. Journal of Food Products Marketing, 27(2), 105-126. doi: 10.1080/10454446.2021.1906817.

Alajali

Zhou

Wen

, & Wang

(2018). Intersection traffic prediction using decision tree models. Symmetry, 10(9), 386.

Alalwan

A.A.

(2020). Mobile food ordering apps: An empirical study of the factors affecting customer e-satisfaction and continued intention to reuse. International Journal of Information Management, 50, 28-44. doi: 10.1016/j.ijinfomgt.2019.04.008.

Alam

M.Z.

Hoque

M.R.

, & Barua

(2020). Factors influencing the adoption of mHealth services in a developing country: A patient-centric study. International Journal of Information Management, 50, 128-143. doi: 10.1016/j.ijinfomgt.2019.04.016.

Alshurideh

Al Kurdi

Salloum

S.A.

Arpaci

, & Al-Emran

(2020). Predicting the actual use of m-learning systems: a comparative approach using PLS-SEM and machine learning algorithms. Interactive Learning Environments, 1-15.

Aureliano-Silva

Leung

, & Spers

E.E.

(2021). The effect of online reviews on restaurant visit intentions: Applying signaling and involvement theories. Journal of Hospitality and Tourism Technology, 12(4), 672-688. doi: 10.1108/JHTT-06-2020-0143.

Bhattacherjee

(2001). Understanding information systems continuance: An expectation-confirmation model. MIS Quarterly, 25(3), 351. doi: 10.2307/3250921.

10.

Cervantes

Garcia-Lamont

Rodríguez-Mazahua

, & Lopez

(2020). A comprehensive survey on support vector machine classification: Applications, challenges and trends. Neurocomputing, 408, 189-215.

11.

Chen

(2008). A model of consumer acceptance of mobile payment. International Journal of Mobile Communications, 6(1), 32. doi: 10.1504/IJMC.2008.015997.

12.

Chen

, & Guestrin

(2016). XGBoost: a scalable tree boosting system. In: Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining ACM, pp. 785-794.

13.

Cho

Bonn

M.A.

, & Li

(Justin). (2019). Differences in perceptions about food delivery apps between single-person and multi-person households. International Journal of Hospitality Management, 77, 108-116. doi: 10.1016/j.ijhm.2018.06.019.

14.

Davis

F.D.

Bagozzi

R.P.

, & Warshaw

P.R.

(1989). User acceptance of computer technology: A comparison of two theoretical models. Management Science, 35(8), 982-1003. doi: 10.1287/mnsc.35.8.982.

15.

Dorogush

A.V.

Ershov

, & Gulin

(2018). CatBoost: gradient boosting with categorical features support. arXiv preprint arXiv:1810.11363.

16.

Dwivedi

A.K.

(2018). Performance evaluation of different machine learning techniques for prediction of heart disease. Neural Computing and Applications, 29(10), 685-693.

17.

Elvandari

C.D.R.

Sukartiko

A.C.

, & Nugrahini

A.D.

(2018). Identification of technical requirement for improving quality of local online food delivery service in yogyakarta. Journal of Industrial and Information Technology in Agriculture, 1(2), 1-7. doi: 10.24198/jiita.v1i2.14573.

18.

Global Finance. (2020). Kuwait Startups: Seeding Tomorrow’s Giants. https://www.gfmag.com/magazine/january-2020/kuwait-startups-seeding-tomorrows-giants.

19.

Gunden

Morosan

, & DeFranco

(2020). Consumers’ intentions to use online food delivery systems in the USA. International Journal of Contemporary Hospitality Management, 32(3), 1325-1345. doi: 10.1108/IJCHM-06-2019-0595.

20.

Hair

J.F.

Hollingsworth

Randolph

, & Chong

(2017). An updated and expanded assessment of PLS-SEM in information systems research. Industrial Management & Data Systems, 117(3), 442-458. doi: 10.1108/IMDS-04-2016-0130.

21.

Hair

J.F.

Risher

J.J.

Sarstedt

, & Ringle

C.M.

(2019). When to use and how to report the results of PLS-SEM. European Business Review, 31(1), 2-24. doi: 10.1108/EBR-11-2018-0203.

22.

Hasan

Palaniappan

Raziff

A.R.A.

Mahmood

, & Sarker

K.U.

(2018, August). Student academic performance prediction by using decision tree algorithm. In 2018 4th International Conference on Computer and Information Sciences (ICCOINS), IEEE, pp. 1-5.

23.

Henrique

B.M.

Sobreiro

V.A.

, & Kimura

(2019). Literature review: Machine learning techniques applied to financial market prediction. Expert Systems with Applications, 124, 226-251.

24.

Henseler

Hubona

, & Ray

P.A.

(2016). Using PLS path modeling in new technology research: Updated guidelines. Industrial Management & Data Systems, 116(1), 2-20. doi: 10.1108/IMDS-09-2015-0382.

25.

Henseler

Ringle

C.M.

, & Sarstedt

(2015). A new criterion for assessing discriminant validity in variance-based structural equation modeling. Journal of the Academy of Marketing Science, 43(1), 115-135. doi: 10.1007/s11747-014-0403-8.

26.

Henseler

Ringle

C.M.

, & Sinkovics

R.R.

(2009). The Use of Partial Least Squares Path Modeling in International Marketing. In R.R. Sinkovics & P.N. Ghauri (Eds.), New Challenges to International Marketing, Emerald Group Publishing Limited, Vol. 20, pp. 277-319. doi: 10.1108/S1474-7979(2009)0000020014.

27.

Hong

, & Jin

(2018). Travel and Online Review Behavior. PACIS 2018 Proceedings. https://aisel.aisnet.org/pacis2018/3.

28.

Huang

Y.-M.

(2019). Examining students’ continued use of desktop services: Perspectives from expectation-confirmation and social influence. Computers in Human Behavior, 96, 23-31. doi: 10.1016/j.chb.2019.02.010.

29.

IMARC. (2020). Global Online Food Delivery Market to Reach US$ 164.5 Billion by 2024, Stimulated by Development of User-Friendly Applications. https://www.imarcgroup.com/global-online-food-delivery-market.

30.

Iwendi

Bashir

A.K.

Peshkar

Sujatha

Chatterjee

J.M.

Pasupuleti

, et al. (2020). COVID-19 patient health prediction using boosted random forest algorithm. Frontiers in Public Health, 8, 357.

31.

Kabari

L.G.

, & Onwuka

U.C.

(2019). Comparison of bagging and voting ensemble machine learning algorithm as a classifier. International Journals of Advanced Research in Computer Science and Software Engineering, 9(3), 19-23.

32.

Kaur

Dhir

Talwar

, & Ghuman

(2021). The value proposition of food delivery apps from the perspective of theory of consumption value. International Journal of Contemporary Hospitality Management, 33(4), 1129-1159. doi: 10.1108/IJCHM-05-2020-0477.

33.

Meng

Finley

Wang

Chen

, & Liu

(2017). LightGBM: a highly efficient gradient boosting decision tree. 31st Conference on Neural Information Processing Systems.

34.

Kim

Galliers

R.D.

Shin

Ryoo

J.-H.

, & Kim

(2012). Factors influencing Internet shopping value and customer repurchase intention. Electronic Commerce Research and Applications, 11(4), 374-387. doi: 10.1016/j.elerap.2012.04.002.

35.

Kim

Mirusmonov

, & Lee

(2010). An empirical examination of factors influencing the intention to use mobile payment. Computers in Human Behavior, 26(3), 310-322. doi: 10.1016/j.chb.2009.10.013.

36.

Kumar

, & Shah

(2021). Revisiting food delivery apps during COVID-19 pandemic? Investigating the role of emotions. Journal of Retailing and Consumer Services, 62, 102595. doi: 10.1016/j.jretconser.2021.102595.

37.

Lai

P.C.

, & Liew

E.J.Y.

(2021). Towards a cashless society: The effects of perceived convenience and security on gamified mobile payment platform adoption. Australasian Journal of Information Systems, 25. doi: 10.3127/ajis.v25i0.2809.

38.

Lee

E.-Y.

Lee

S.-B.

, & Jeon

Y.J.J.

(2017). Factors influencing the behavioral intention to use food delivery apps. Social Behavior and Personality: An International Journal, 45(9), 1461-1473. doi: 10.2224/sbp.6185.

39.

Lee

S.W.

Sung

H.J.

, & Jeon

H.M.

(2019). Determinants of continuous intention on food delivery apps: Extending UTAUT2 with information quality. Sustainability, 11(11), 3141. doi: 10.3390/su11113141.

40.

, & Mao

(2014). A review of boosting methods for imbalanced data classification. Pattern Analysis and Applications, 17(4), 679-693.

41.

Maimaiti

Zhao

Jia

, & Zhu

(2018). How we eat determines what we become: Opportunities and challenges brought by food delivery industry in a changing world in China. European Journal of Clinical Nutrition, 72(9), 1282-1286. doi: 10.1038/s41430-018-0191-1.

42.

Marinković

Dordević

, & Kalinić

(2020). The moderating effects of gender on customer satisfaction and continuance intention in mobile commerce: A UTAUT-based perspective. Technology Analysis & Strategic Management, 32(3), 306-318. doi: 10.1080/09537325.2019.1655537.

43.

Mehrolia

Alagarsamy

, & Solaikutty

V.M.

(2021). Customers response to online food delivery services during COVID-19 outbreak using binary logistic regression. International Journal of Consumer Studies, 45(3), 396-408. doi: 10.1111/ijcs.12630.

44.

Okumus

, & Bilgihan

(2014). Proposing a model to test smartphone users’ intention to use smart applications when ordering food in restaurants. Journal of Hospitality and Tourism Technology, 5(1), 31-49. doi: 10.1108/JHTT-01-2013-0003.

45.

Pham

T.-T.T.

, & Ho

J.C.

(2015). The effects of product-related, personal-related factors and attractiveness of alternatives on consumer adoption of NFC-based mobile payments. Technology in Society, 43, 159-172. doi: 10.1016/j.techsoc.2015.05.004.

46.

Qahri-Saremi

, & Montazemi

A.R.

(2022). Negativity bias in the diagnosticity of online review content: The effects of consumers’ prior experience and need for cognition. European Journal of Information Systems, 0(0), 1-18. doi: 10.1080/0960085X.2022.2041372.

47.

Qian

Wang

Yuan

Gao

, & Song

(2022). Financial distress prediction using a corrected feature selection measure and gradient boosted decision tree. Expert Systems with Applications, 190, 116202.

48.

Rabaa’i

A.A.

(in press a). An Investigation into the Acceptance of Mobile Wallets in the FinTech Era: An Empirical Study from Kuwait. International Journal of Business Information Systems. doi: 10.1504/IJBIS.2021.10038422.

49.

Rabaa’i

A.A.

(in press b). Factors Affecting the Adoption of Mobile Grocery Delivery Applications (MGDAs). International Journal of Business Information Systems. doi: 10.1504/IJBIS.2021.10045250.

50.

Rabaa’i

A.A.

(in press c). What’s for Dinner? Factors Contributing to the Continuous Usage of Food Delivery Apps (FDAs). Asia Pacific Journal of Information Systems.

51.

Rabaa’i

A.A.

, & Abu ALMaati

(2021). Exploring the determinants of users’ continuance intention to use mobile banking services in kuwait: Extending the expectation-confirmation model. Asia Pacific Journal of Information Systems, 3, 141-184. doi: 10.14329/apjis.2021.31.2.141.

52.

Rabaa’i

A.A.

Al-lozi

Hammouri

Bany Muhammad

Alsmadi

, & Al-Gasawneh

(2022). Continuance intention to use smartwatches: An empirical study. International Journal of Data and Network Science, 6. doi: 10.5267/j.ijdns.2022.4.012.

53.

Rabaa’i

A.A.

ALmaati

S.A.

, & Zhu

(2021). Students’ continuance intention to use moodle: An expectation-confirmation model approach. Interdisciplinary Journal of Information, Knowledge, and Management, 16, 397-434. doi: 10.28945/4842.

54.

Rabaa’i

A.A.

Bhat

, & Abu Al Maati

(2018). Theorising social networks addiction: An empirical investigation. International Journal of Social Media and Interactive Learning Environments, 6(1), 1. doi: 10.1504/IJSMILE.2018.10013518.

55.

Rabaa’i

A.A.

, & Zhu

(2021). Understanding the determinants of wearable payment adoption: An empirical study. Interdisciplinary Journal of Information, Knowledge, and Management, 16, 173-211. doi: 10.28945/4746.

56.

Ray

Dhir

Bala

P.K.

, & Kaur

(2019). Why do people use food delivery apps (FDA)? A uses and gratification theory perspective. Journal of Retailing and Consumer Services, 51, 221-230. doi: 10.1016/j.jretconser.2019.05.025.

57.

Ringle

Wende

, & Becker

(2015). SmartPLS 3. Boenningstedt: SmartPLS GmbH. http://www.smartpls.com.

58.

Rogers

(2003). Diffusion of Innovations (5th ed.). Free Press.

59.

Roh

, & Park

(2019). Adoption of O2O food delivery services in South Korea: The moderating role of moral obligation in meal preparation. International Journal of Information Management, 47, 262-273. doi: 10.1016/j.ijinfomgt.2018.09.017.

60.

Sahu

S.K.

Prasad

M.B.N.V.

, & Tripathy

B.K.

(2015). A support vector machine binary classification and image segmentation of remote sensing data of Chilika Lagloon. International Journal of Research in Information Technology, 3(5), 191-204.

61.

Sarker

I.H.

Kayes

A.S.M.

, & Watters

(2019). Effectiveness analysis of machine learning classification models for predicting personalized context-aware smartphone usage. Journal of Big Data, 6(1), 1-28.

62.

Sarstedt

Ringle

C.M.

Smith

Reams

, & Hair

J.F.

(2014). Partial least squares structural equation modeling (PLS-SEM): A useful tool for family business researchers. Journal of Family Business Strategy, 5(1), 105-115. doi: 10.1016/j.jfbs.2014.01.002.

63.

Speiser

J.L.

Miller

M.E.

Tooze

, & Ip

(2019). A comparison of random forest variable selection methods for classification prediction modeling. Expert Systems with Applications, 134, 93-101.

64.

Statistia. (2020). eServices Report 2019 – Online Food Delivery. https://www.statista.com/study/40457/food-delivery/.

65.

Suhartanto

Helmi Ali

Tan

K.H.

Sjahroeddin

, & Kusdibyo

(2019). Loyalty toward online food delivery service: The role of e-service quality and food quality. Journal of Foodservice Business Research, 22(1), 81-97. doi: 10.1080/15378020.2018.1546076.

66.

Talabat. (2021). The number of restaurants in Kuwait. https://www.talabat.com/kuwait/restaurants?page=121.

67.

Tanha

Abdi

Samadi

Razzaghi

, & Asadpour

(2020). Boosting methods for multi-class imbalanced data classification: An experimental review. Journal of Big Data, 7(1), 1-47.

68.

Troise

O’Driscoll

Tani

, & Prisco

(2020). Online food delivery services and behavioural intention – a test of an integrated TAM and TPB framework. British Food Journal, 123(2), 664-683. doi: 10.1108/BFJ-05-2020-0418.

69.

Uddin

Khan

Hossain

M.E.

, & Moni

M.A.

(2019). Comparing different supervised machine learning algorithms for disease prediction. BMC Medical Informatics and Decision Making, 19(1), 1-16.

70.

Venkatesh

Morris

M.G.

Davis

G.B.

, & Davis

F.D.

(2003). User acceptance of information technology: Toward a unified view. MIS Quarterly, 27(3), 425-478. JSTOR. doi: 10.2307/30036540.

71.

Venkatesh

Thong

J.Y.L.

, & Xu

(2012). Consumer acceptance and use of information technology: Extending the unified theory of acceptance and use of technology. MIS Quarterly, 36(1), 157-178. JSTOR. doi: 10.2307/41410412.

72.

Wang

W.-T.

W.-M.

, & Chen

W.-Y.

(2019). The impact of inertia and user satisfaction on the continuance intentions to use mobile communication applications: A mobile service quality perspective. International Journal of Information Management, 44, 178-193. doi: 10.1016/j.ijinfomgt.2018.10.011.

73.

Yen

C.-L.A.

, & Tang

C.-H.H.

(2019). The effects of hotel attribute performance on electronic word-of-mouth (eWOM) behaviors. International Journal of Hospitality Management, 76, 9-18. doi: 10.1016/j.ijhm.2018.03.006.

74.

Yeo

V.C.S.

Goh

S.-K.

, & Rezaei

(2017). Consumer experiences, attitude and behavioral intention toward online food delivery (OFD) services. Journal of Retailing and Consumer Services, 35, 150-162. doi: 10.1016/j.jretconser.2016.12.013.

75.

Zawya. (2021). Kuwait eateries, cafes climb in count. https://www.zawya.com/mena/en/business/story/Kuwait_eateries_cafes_climb_in_count-SNG_244929858/.

76.

Zhao

, & Bacao

(2020). What factors determining customer continuingly using food delivery apps during 2019 novel coronavirus pandemic period? International Journal of Hospitality Management, 91, 102683. doi: 10.1016/j.ijhm.2020.102683.

77.

Zien

Krämer

Sonnenburg

, & Rätsch

(2009, September). The feature importance ranking measure. In Joint European Conference on Machine Learning and Knowledge Discovery in Databases, Springer, Berlin, Heidelberg, pp. 694-709.

The use of machine learning to predict the main factors that influence the continuous usage of mobile food delivery apps

Abstract

Keywords

1. Introduction

2. Theoretical foundation and the constructs used in this study

2.1 Convenience (CON)

2.2 Perceived compatibility (PC)

2.3 Online reviews (OR)

2.4 Delivery experience (DE)

3. Methodology

3.1 Measurement instrument

3.2 Data collection and sample

Table 1 Demographic characteristics of respondents

4.1 Reliability and validity of the measurement scale

Table 2 Items loading, Cronbach’s alpha, composite reliability, and AVE

Table 4 Prediction results using machine learning models

6. Conclusion, limitations and future research

References

Table 1
Demographic characteristics of respondents

Table 2
Items loading, Cronbach’s alpha, composite reliability, and AVE

Table 4
Prediction results using machine learning models