Hybrid fuzzy logic with SVM based prediction analysis model to predict innovation performance of 3C Industry

Abstract

This paper explores the applications of support vector machines (SVM) technique and panel data econometric approach in innovation performance. We proposed Hybrid Fuzzy Logic with SVM based prediction analysis model to predict Innovation Performance of 3C Industry and then we construct the top management innovation awareness by text mining analysis, which differs from traditional methods, and then discusses the mediation function of financial flexibility in the relationship between innovation awareness and enterprise performance. Then, we use the financial econometric method panel regression and apply the SVM, a statistical technique that has gained special popularity in the field of AI to test the relation between innovation performance and innovation awareness of top management in 3C industry. The findings show that there exists a significant position relationship between innovation awareness and innovation performance, and that the mediation function of financial flexibility does work. With the SVM approach, the innovation performance can be predicted well by top management innovation awareness.

Keywords

Support vector machines text mining innovation performance innovation awareness financial flexibility

1 Introduction

With the emergence of innovative technologies in the field of information technology the one more innovative concept referred to as 3C industry is increasingextremely. Innovation has been the focus among nations, enterprises, and scholars around the world in recent years. 3C is the general term for computers, communications and consumer electronics products. Technological innovation has become a new force for 3C industry manufacturers, and at the same time, the innovation awareness of top management takes an important part for the company’s development. Recently, there are two streams to model the corporate’s innovation performance which are the traditional econometric approach and modern artificial intelligent approach. In this study, we apply the hybrid model based on concept of Fuzzy logic and support vector machines to verify the importance of top management innovation awareness and to predict the innovation performance.

Support vector machine can build a hyperplane in differentinterplanetary to split points into two different plane. It is not tough to visualize that a best classifier resolute by hyperplane is the one with the major distance from the nearby training dataset points in these two categories. SVM is very ideal in many applications because of its better accuracy prediction behavior and universaloptimumresult can be achieved. SVM always pursues hyperplane as far-off from the nearby dataset point as possible on individually side which can solve the linear separable problem.

3C is the general term for computers, communications and consumer electronics products. Technological innovation has become a new force for 3C industry manufacturers, and at the same time, the innovation awareness of top management takes an important part for the company’s development. Therefore, we use the technique of text mining analysis to explore the innovation awareness of the top management of company.

The new economic theory indicates that innovation is the core power of economic growth, whereas the enterprise is the subject of innovation. The best way to realize sustainable economic development is to increase enterprise efficiency via innovation. The main contribution of this work is as follows:

(1) We proposed Hybrid Fuzzy Logic with SVM based prediction analysis model to predict Innovation Performance of 3C Industry

(2) We construct the top management innovation awareness by text mining analysis

(3) We use the financial econometric method panel regression to test the relation between innovation performance and innovation awareness of top management in 3C industry

The remainder of paper is structured as follows: section 2 gives brief overview of literature review related to this work. Methodology and dataset description is describing in section 3. Section 4 represent the empirical analysis of proposed system with result outcomes and performance evaluation. The whole work is summarizing with final concluding remark and future recommendation in section 5.

2 Related work

Innovation needs research and development (R&D) investments, which demand excessive and long-term capital support. If a company lacks stable financing resources to maintain R&D, then it will be difficult for innovation performance to be realized even if firm management exhibits strong innovation awareness. The top management team is the management group responsible for enterprise strategy planning and decision-making. Top management team members work together and have authority over deciding the direction of their firm’s future development [17]. Management innovation awareness refers to a kind of cognition, manner, and willingness to execute that help excavate an enterprise’s innovation potential by using enterprise resources for innovation activity.

The literature on management awareness mostly relates to corporate development strategy. In [7] Porac et al. found that top management’s cognition of the external environment influences the organization’s decision, planning, and structure. In [4] Johnson et al. pointed out the mediation function of management awareness, such as external environment and strategy selection, degree of industry competition, and strategy.

Many scholars have found that financing flexibility has an inhibiting effect on an enterprise’s R&D investments, which can be improved by maintaining higher financial flexibility. In [9] Goodale, et al. also report that the management behaviors affect the corporate’s innovation performance. In [2] Heavey et al. concern about the top management taking an important part on company’s innovation.

In [13] Benhayoun, et al. use the SVM to explore the financial indicators’ ability of perdition of a firm’s credit. Therefore, this paper discusses the relationship between financial flexibility and innovation as well as the mediating effects.

Meanwhile, support vector machine (SVM) is a one of machine learning algorithmsin the AI field for regression and become very widespread and employed in several applications [11, 14]. We examine the relation between innovation performance and innovation awareness of top management in 3C industry with SVM.

The amalgamation of more than one computational based Intelligence structures as a distinctexemplary is known as Hybrid Computational based Intelligence (HCI) and it is attractivegradually as well as very popular in recent research studies. This amplifiedattractiveness lies in the widespreadachievement of hybrid structures in numerouspracticalintricatecomplications [6]. Animportantcondition for the inclusion of technologies is the presence of a “shared denominator” to construct upon [19]. In this work, we use Fuzzy system with Support Vector Machines (SVM) hybridization for exceptional predictive abilities.

3 Methodology used

In this section first we proposed method based computer intellect based hybrid approach of artificial intelligence that include the concept of fuzzy logic(FL) and support vector machine (SVM). Then we construct the top management innovation awareness by text mining analysis, which differs from traditional methods, and then discusses the mediation function of financial flexibility in the relationship between innovation awareness and enterprise performance.

3.1 FL-SVM prediction analysis model

In this section we briefly describe FL-SVM Prediction Analysis Model and the framework architecture is mention in Fig. 1. We use hybridization technique in this work in order to extract strength of each individual technique at the same time suppress the weakness of one technique with the strength of another one.

Fig.1

Framework Hybrid Fuzzy Logic with SVM based prediction analysis model to predict Innovation Performance of 3C Industry (a) FL-SVM Prediction Analysis Model during training phase. (b) FL-SVM Prediction Analysis Model during testing phase.

3.1.1 Stratified sampling

First we divide input dataset into training and testing dataset using standard Stratified sampling method [1, 17]. This method divides the dataset into 70 : 30 ratios that means 70% dataset is used during training phase while remaining 30% is used for testing phase.

3.1.2 Fuzzy logic system

The training dataset then first passess to the Fuzzy logic system that process the data using fuzzy rule base knowledge and interferencing mechanism [5].

3.1.2.1 Fuzzification

Fuzzification is the process of converting crisp data input into fuzzy linguistic variable in the form of membership function [8]. Let A be a universal nonempty set, which consider universal discourse. $A = {a_{1}, a_{2}, \dots a_{n}$ (1)

A fuzzy set F is defined under universal discourse of nonempty set A ={ a₁, a₂, … a_n } with an ordered pair set represented as: ${(a_{1}, μ_{F} (a_{1})), (a_{2}, μ_{F} (a_{2})) \dots (a_{n}, μ_{F} (a_{n}))}$ (2) and is characterized by degree of membership function μ_F (A) that maps every element a in A to real number in the interval 0,1. Fuzzy membership function evaluation equation is described using following equation: $μ_{IN} (x, a_{1}, a_{2}, a_{3}) = {\begin{matrix} 0, & if x ⩽ a_{1}; \\ \frac{x - a_{1}}{a_{3} - a_{1}}, & if a_{1} < x < a_{3}; \\ \frac{a_{2} - x}{a_{2} - a_{3}}, & if a_{3} < x < a_{2}; \\ 0, & if x ⩾ a_{2} \end{matrix}$ (3)

Where, x represent membership degree and p, q, r represent leftmost, middle and rightmost membership value in triangular fuzzy number,. Each input value of all input variables is converted into fuzzy input sets, define as: μ_IN (x, a₁, a₂, a₃) ={ High (H) , Medium (M) , Low (L) , VeryLow (VL) }. All input variable is represent based on Gaussian-model membership functions.

3.1.2.2 Fuzzy rule base interference engine

This module performs min-max operation over the inputs received. We perform Min-max operation two time during the procedure, first we use service industry parameters and perform min-max operation with Mamdani rule validation [13]. Within fuzzy interferenceengine we use min-max operation using following equation: $ρ_{j} = m i n_{k = 1}^{m} [\max_{X_{k}} (μ_{k}^{'} (x_{k}) \land μ_{j k} (x_{k}))]$ (4)

Where, represent new autonomous values of input variable. Here, we use Mamdani rule implication that generate individual fuzzy output in the form of membership function [16]. The general format of rule implication is as follows: $\begin{matrix} IF x_{1} is μ_{11} and \dots \dots x_{m} is μ_{1 m} THEN y is {IO}_{1}; \\ IF x_{1} is μ_{m 1} and \dots \dots x_{m} is μ_{nm} THEN y is {IO}_{m}; \end{matrix}$

Where x_k = 1 … m represent input and output sets.

3.1.2.3 Defuzzification

The output received is in fuzzy form which is converted into real crisp values using following equation of center of gravity (COG) of defuzzification method: $y = \int_{sp} IO (y) ydy / \int_{sp} IO (y) dy$ (5)

The Fuzzy logic system then finally produce the crisp output which is known as intermediate output (IO), as fuzzy system has an capability to processess uncertain data that’s why first we passes the input dataset into fuzzy system that product intermediate output [17]. This intermediate output is then combined with training dataset then passes to SVM prediction analysis model to produce final prediction outcome.

3.1.3 Support vector machine (SVM) approach

With a given set of training+intermediate output dataset SVM optimize the problem with input pair (p_f, q_f) where k = 1, 2, 3, … . n ; p_f ∈ Randq_f ∈ { - 1, 1 } ⁿ The formation of binary SVM cost function is defined by following equation: $\begin{matrix} \min_{x, y, ɛ} \frac{1}{2} x^{T} x + ∁_{r} \sum_{f} ɛ_{f} \\ subject to q_{f} . (x^{T} ø (p_{f}) + y) ⩾ 1 - ɛ_{f}, \\ ɛ_{f} ⩾ 0 . \end{matrix}$ (6)

Where, x and y represent separating hyper plane parameter, ɛ_f represent penalty error for loose variable (located at erroneous side of hyper plane margin), _∁r represent control parameter for regularization that trade-off between hyper plane margin and penalty error ɛ_f, ø (p_f) represent input vector non-linear transformation function. The two separating hyper plane can be defined as x^T ø (p_f) + y = 1 and x^T ø (p_f) + y = 0, while margin width is define as 2/∥ x ∥ . We use the radial basis function (RBF) kernel is used for SVM classifier which is defined by following by following equation: $K (p_{f}, p_{g}) \equiv ø {(p_{f})}^{T} ø (p_{g}) = e^{- γ p_{f} - p_{g}^{2}}$ (7)

SVM usually can obtain more accurate global optimal solution and prediction. Generally, RBF kernel is a reasonable first choice which maps the sample nonlinearity to a high dimensional space, so it is different from the linear kernel and can deal with the case that the relationship between class tags and attributes is nonlinear.

3.2 Text mining of innovation awareness of top management

Innovation awareness of TMT refers to its acknowledgement toward innovation, including the degree of emphasis, the willingness to carry out innovation activity, as well as the forecasting of innovation performance for the whole enterprise. Scholars have conducts questionnaire surveys or psychology tests to obtain data related to this awareness, views on value, and some other features, but this method is not easy and could lead to difficulty of data collection. Research nowadays mainly measures the awareness pattern through top management’s demographic characteristics, but inaccuracies have been found here, too, as shown by Damanpour and Schneider (2006). Therefore, our study employs a text analysis method that is based on the Sa-pir-Whorf hypothesis (1944). According to Whorf, different cultures and cognitive styles are coded in different languages that influence people’s thinking. As a result, thoughts and appearances differ on account of the distinction in the language that people use. In short, language is a reflection of awareness.

We first select the yearly reports of the targeted enterprises from the CNINFO website and change the format from PDF to TXT. Second, we design a program in Python to calculate the number of key words, including “innovation”, “autonomously”, “research and development”, “scientific research”, “patent”, “invention”, “research”, “new product”, “intellectual property rights”, and ten other words that may denote expression of top management’s awareness. We define innovation as the number of licensed invention patents divided by R&D investment the year before. Patents cover both invention and design, but invention reflects innovation better. After counting the number of total words in the yearly reports, we divide the numbers of key words by it to get the portion that denotes the degree of top management innovation. We utilize JIEBA lexicons for the Chinese words’ segmentation in the text analysis.

3.3 Dataset description

Our sample contains the Chinese A-share firms which is categorized under the Chinese Securities Regulatory Commission industrial category C39 (computer, communications and other electronic equipment manufacturing enterprises). To make sure the variable is available, we exclude the enterprises which is (1) ST, *ST company; (2) lacking samples seriously; (3) not publishing the R&D investments. The sample period spans from 2011 to 2016. The data mainly comes from CSMAR and WIND database, while the year reports used for text analysis are from CNINFO web.

The variables are as follows.

Innovation Performance (INNO): number of inventions of the company.

Innovation awareness (IA): number of key words / number of total words in yearly report.

Average education level (EDU): scores 1–5 stand for secondary school, junior college, bachelor, master, and doctor; from the average score, if the level < 3.5, then EDU = 0; if the level > = 3.5, then EDU = 1.

Average TMT age (AGE): if average age < 50, then AGE = 0; if average age > = 50, then AGE = 1.

Female TMT proportion (FEMALE): the proportion of females in TMT.

Average TMT tenure (TENURE): the ln value of average TMT tenure.

Financial flexibility (FF): financial flexibility reserve equals cash flexibility plus debt flexibility; FF = (cash ratio of enterprise – average cash ratio of industry) +MAX (0, average debt ratio of industry – debt ratio of enterprise).

Growth (GROW): growth rate of sales revenue.

State-owned enterprise (NATION): if the enterprise is state-owned, then NATION = 0; if the enterprise is not state-owned, then NATION = 1.

Management share (SHARE): the ratio of executives’ shareholdings.

4 Empirical results

4.1 Traditional econometric method-panel regression with principal components approach

The F test and Hausman test suggest using panel regression with random effect. So, we select the cross-section random panel regression to run the formulation, the result is as followed.

Table 1 shows the panel regression results of innovation awareness and innovation performance. Here, EDU shows a significant positive relationship with Innovation Performance, which means a higher level of education increases innovation performance.

Table 1
Innovation awareness and innovation performance

Dependent Variable: Innovation Performance

Variable Coefficient t-Statistic Prob.

IA –0.0132 –0.0323 0.9742

EDU 2.8201^*** 3.4598 0.0006

FEMALE –2.6071 –0.6410 0.5220

TENURE 0.3212 0.7661 0.4442

AGE –0.2601 –0.3003 0.7641

NATION –1.1564 –1.6699 0.0959

GROW 0.3155 0.3981 0.6908

SHARE 0.2149 0.9396 0.3481

C 1.4390 0.5979 0.5503

R-squared 0.0472 F-statistic 1.9514

Adjusted R-squared 0.0230 S.E. of regression 3.6823

Dependent Variable: Innovation Performance
IA	–0.0132	–0.0323	0.9742
EDU	2.8201^***	3.4598	0.0006
FEMALE	–2.6071	–0.6410	0.5220
TENURE	0.3212	0.7661	0.4442
AGE	–0.2601	–0.3003	0.7641
NATION	–1.1564	–1.6699	0.0959
GROW	0.3155	0.3981	0.6908
SHARE	0.2149	0.9396	0.3481
C	1.4390	0.5979	0.5503
R-squared	0.0472	F-statistic	1.9514
Adjusted R-squared	0.0230	S.E. of regression	3.6823

We next use principle component analysis (PCA) to construct a factor, including CONS, EDU, AGE, FEMALE, and TUNURE, to represent top management’s innovation awareness (TIA). Table 2 shows the principle component coefficient matrix of innovation awareness.

Table 2

Principle component coefficient matrix of top management’s innovation awareness

Eigenvectors (loadings):
Variable	PC 1	PC 2	PC 3	PC 4	PC 5
IA	0.052406	0.715218	–0.569465	–0.36499	0.167957
EDU	0.472877	0.329964	–0.092028	0.802921	–0.119831
FEMALE	–0.254084	0.600442	0.676458	–0.06963	–0.335361
TUNURE	0.578249	0.060523	0.456825	–0.218003	0.636986
AGE	–0.612137	0.124072	0.030909	0.411978	0.662732

Table 3 shows the relationship among innovation awareness, innovation performance, and corporate performance. Top management’s innovation awareness being significant shows that TMT takes awareness, age, education, gender, and tenure into consideration when enhancing its firm’s innovation performance. This supports the first hypothesis. The coefficient of NATION is significantly below zero, indicating non-state-owned enterprises do better at innovation activity. Private enterprises attach more importance to R&D and invention. As to SHARE, it has a significantly negative effect on INNO. According to Jensen and Meckling (1976), when the ratio of shares held by management rises, the interests of the two are more aligned. Many studies have found that increasing this phenomenon results in the entrenchment hypothesis on investment decision-making. They have more consideration in R&D investment, finally leading to a negative relationship with innovation performance.

Table 3

Innovation awareness, innovation performance, and corporate performance

Dependent Variable: Innovation Performance
Variable	Coefficient	Std. Error	Prob.
TIA	4.181327^***	1.333254	0.0018
NATION	–0.950588^*	0.561363	0.0912
GROW	0.152738	0.608475	0.8019
SHARE	–1.186232^***	0.375423	0.0017
C	2.203736	1.192312	0.0653
R-squared	0.032292	F-statistic	3.153474
Adjusted	0.022052	Durbin-Watson	0.822091
R-squared		stat

We regress TIA on the mediation variable FF, obtaining a significant coefficient “a” that equals 1.106. A higher value of the series leads to higher FF. Looking Table 3 the coefficients “TIA” and “FF ” are both significant; according to the fourth step, some other mediation variables besides FF also work. The masking effect does not appear, and FF has a partial mediation effect on the relationship between ROA and TIA, taking up to $\frac{1.106 * 0.005}{0.050} = 11.42 %$ in the total effect.

4.2 FL-SVM simulation for innovation performance prediction

We select C-SVM which is an application to regress and classify data with the radial basis function and introduced by [3]. Figure 2 shows the C-SVM selection of the best parameter of c and g for the prediction of innovation performance by using TIA and FF. To demonstrate the efficacy of prediction, we compare FL-SVM with C-SVM model. Table 4 shows the prediction results of SVM. Fig. 3 shows the MSE of C-SVM simulation for innovation performance prediction. The results show that TIA and FF predict the corporate’s innovation performance with SVM model which provides good precisions.

Fig.2

C-SVM selection of the best parameter of c and g.

Fig.3

MSE of C-SVM selection of simulation for innovation performance prediction.

SVM gives intelligent suggestionsand shows the importance of the innovation awareness of top managers.

5 Conclusion

This research predicts the corporate’s innovation performance with FL-SVM based prediction analysis model. The idea of our research is derived from the fact that in the process of text mining of top management innovation with the combination of features of corporate financial performance datasets might have bettersignificantimpact than customaryparticular factor. We authenticated our proposed model with actual data sets, and improved the accuracy of prediction with reduced learning time of SVM models.

Innovation awareness of top management, financial flexibility, and corporate performance are important issues in modern financial economics. In this paper we illustrate these ideas and implement techniques with traditional econometric method such as panel regression and principle component analysis to confirm the innovation awareness of top management derived from text-mining approach has a positive effect on innovation performance and a positive effect on enterprise performance. Then, we force on the prediction of SVM to provide intelligent suggestions and shows the importance of the innovation awareness of top managers from real financial data.

This study offers important implications for top corporate managers concerning about how to balance innovation and financial flexibility in their enterprise. The resultsprovide a brand new analysis of the relation between innovation performance and innovation awareness of top management in 3C industry which may help the investors, managers and investigators of financial institutions realize the importance of innovation.

Footnotes

Acknowledgments

This work was supported by the Social Science Fund of Fujian Province, China under Grant number FJ2018B075 and Fuzhou University of International Studies and Trade under Grant number FYKQJ201904 and Grant number FYKQJ202003.

References

Anifowose

F.A.

and Abdulraheem

, Prediction of porosity and permeability of oil and gas reservoirs using hybrid computational intelligence models, In North Africa Technical Conference and Exhibition, Society of Petroleum Engineers, (2010).

Heavey

and Simsek

, Top Management Compositional Effects on Corporate Entrepreneurship: The Moderating Role of Perceived Technological Uncertainty, Journal of Product Innovation Management 30(5) (2013), 837–855.

Chang

C.-C.

and Lin

C.-J.

, LIBSVM: a Library for Support Vector Machines, ACM Transactions on Intelligent Systems and Technology 2(27) (2011), 1–27.

Johnson

D.R.

and Hoopes

D.G.

, Managerial Cognition, Sunk Costs, and the Evolution of Industry Structure, Strategic Management Journal 24(10) (2003), 1057–1068.

Devi

S.S.

, Singh

N.H.

and Laskar

R.H.

, Fuzzy C-Means Clustering with Histogram based Cluster Selection for Skin Lesion Segmentation using Non-Dermoscopic Images, International Journal of Interactive Multimedia and Artificial Intelligence 6(Special Issue on Soft Computing) (2020), 26–31.

Giovanni

and Vincenzo

, Using FML and Fuzzy technology in adaptive ambient intelligence environments, International Journal of Computational Intelligence Research 1(2) (2005), 171–182.

Helmy

, Fatai

and Faisal

, Hybrid computational models for the characterization of oil and gas reservoirs, Expert Systems with Applications 37(7) (2010), 5353–5363.

Ibrahim

, Kamel

, Rashad

, Nasrat

and Jurado

, Performance Enhancement of Wind Farms Using Tuned SSSC Based on Artificial Neural Network, International Journal of Interactive Multimedia and Artificial Intelligence 5(Regular Issue) (2019), 118–124.

Goodale

J.C.

, Kuratko

D.F.

, Hornsby

J.S.

and Covin

J.G.

, Operations Management and Corporate Entrepreneurship: The Moderating Effect of Operations Control on the Antecedents of Corporate Entrepreneurial Activity in Relation to Innovation Performance, Journal of Operations Management 29(1-2) (2011), 116–127.

10.

Porac

J.F.

, Thomas

and Baden-Fuller

, Competitive Groups as Cognitive Communities: The Case of Scottish Knitwear Manufacturers, Journal of Management Studies 26(4) (1989), 397–416.

11.

Jensen

M.C.

and Meckling

W.H.

, Theory of the Firm: Managerial Behavior, Agency Costs, and Ownership Structure, Journal of Financial Economics 3(4) (1976), 305–360.

12.

Poongodi

, Sharma

, Vijayakumar

, Bhardwaj

, Sharma

A.P.

, Iqbal

and Kumar

, Prediction of the Price of EthereumBlockchain Cryptocurrency in an Industrial Finance System, Computers & Electrical Engineering 81 (2020), 106527.

13.

Manickam

and Devarasan

, Level 2 feature extraction for latent fingerprint enhancement and matching using type-2 intuitionistic fuzzy set, International Journal of Bioinformatics Research and Applications 15(1) (2019), 33–50.

14.

Benhayoun

, Chairi

, El Gonnouni

and Lyhyaoui

, Financial Intelligence in Prediction of Firm’s Creditworthiness Risk: Evidence from Support Vector Machine Approach, Procedia Economics and Finance 5 (2013), 103–112.

15.

Ghezelbash

, Maghsoudi

and Carranza

E.J.M.

, Performance Evaluation of RBF-and SVM-Based Machine Learning Algorithms for Predictive Mineral Prospectivity Modeling: Integration of SA Multifractal Model And Mineralization Controls, Earth Science Informatics 12(3) (2019), 277–293.

16.

Ngan

R.T.

, Ali

, Fujita

, Giang

N.L.

, Manogaran

and Priyan

M.K.

, A New Representation of Intuitionistic Fuzzy Systems and Their Applications in Critical Decision Making, IEEE Intelligent Systems..

17.

Rodríguez

, Corredor

and Suárez

, Genetic Operators Applied to Symmetric Cryptography, International Journal of Interactive Multimedia and Artificial Intelligence 5(Regular Issue) (2019), 39–49.

18.

Finkelstein

, Hambrick

D.C.

, Strategic Leadership, West Publishing Company, (1996).

19.

Symeonidis

L.A.

and Mitkas

A.P.

, Agent Intelligence through Data Mining, Multiagent Systems, Artificial Societies, and Simulated Organizations Series, In: International Book Series 14 (2005).