Regression analysis based decision support system with relationship extraction

1. Introduction

Data analytics techniques are playing crucial roles in all types of businesses to draw efficient decisions in the result analysis phases. These techniques provide huge insights and meaningful relationships between datasets (Razia et al.,2017). Multiple data processing methods are originated from these techniques. Many supervised and unsupervised learning algorithms are being used in industries for analysis purposes. Among them, regression techniques are considered the widely used method to extract the relationships between two variables.

2. Types of regression

Regression analysis techniques are classified into three types such as linear, logistic, and multiple linear regression for multiple values. Linear regression is employed to deal with single dependent and single independent variables. Whereas logistic regression is applied for predicting binary dependant variables. On the other hand, multiple linear regression is applied to cope with more than two independent variables. The following Table 1 clearly explains the types of regression analysis.

OPEN IN VIEWER

Many machine learning algorithms have been proposed for disease analysis and predictions (Dudi et al., 2019). But the ML algorithms are used for enhancing the DSS. The decision support system of any organization produces the expected and predicted information from the hidden data. In this proposed work, an institution’s worker data set has been applied to DSS for drawing the final conclusions. Many stochastic regression models have been implemented so far were failed to produce the expected results (Patel et al., 2019).

This DSS has many phases to process the given dataset such as names of the workers, employee ID, years of experience, target achieved, and performance metrics.

4. Salary prediction methods

The salary prediction process also involves the promotional statements as well as the hike to employees. As per the Fig. 2 the employee data has been collected as 115 samples.

OPEN IN VIEWER

Initially, the correlation analysis is performed on the given dataset. Further to this, the strength of association between variables is carried out (Kawahara et al., 2020). This implementation involves the initial phase of importing the required libraries to apply the data pre-processing. It involves the removal of repeated data, missing values, and unappropriated data. NumPy array is used to analyse the data and with the help of panda’s library, the given dataset is imported to the implementation environment (Donthi et al., 2019).

The given dataset is separated as training set and testing set. For training 80% of the dataset is considered whereas for testing the remaining are considered.

6. Applying the model to fit on the data

With help of the linear regression model, the dataset is employed to fit in the conditions. Initially, a single independent variable is considered for the prediction of a single dependent variable as shown in the following Table 2.

As per the Tables 1 and 2 the dependent variable is salary, and the independent variable is years of experience. So, this is called the single linear regression model. In case of two or more independent variables are considered, that is called multiple linear regression. The likelihood relationship of these variables is found clearly (Satish Kumar et al., 2019).

OPEN IN VIEWER

As per the results shown in Fig. 3, the correlation coefficient has been carried out from dependent and independent variables. Based on the performance and years of experience the salary is increased. These visual-based results are clarifying the relationship between the employees’ salaries. The linear and multiple regression are purely different from other available regression types.

In logistic regression, the categorical values as binary values of the variable are predicted with help of the fitting models. In contrast to that, the polynomial regression is applied and implemented on the n th degree polynomial function in the feature space. The linear equation takes some input values to be assumed in the X and Y axis accordingly. The response variable is fitted in the X-axis and the predictor variable is fitted in the Y-axis. Based on the different amounts of data measurement and changes the regression line is plotted. To deal with the exhaustive exploration application with huge data, nonlinear regression models are used (Mahaboob et al. 2019). Due to the errors found in the stochastic regression model, those models are considered inefficient (Praveen et al., 2019).

8. Conclusion

In this proposed work, the widely used regression analysis techniques and their types are implemented on 115 sample datasets. The results are carried out about the relationship and correlation coefficient parameters. In the meanwhile, multiple regression lines are plotted to explain the results in feature space. Due to the inefficiency of classification algorithms and results, in the future, advanced algorithms will be used (Donthi et al., 2019) These ML-based classification algorithms are being used in the healthcare domain in recent days. Hence, as future work, these regression algorithms will be employed in the healthcare domain also (Saba et al., 2020).

9. Future work

The employees of other department salaries also considered in the future work. The logistic regression will be implemented to find the relationship between the entry and exit of the employees. With help of the support vector, machine-based regression models will be used to predict the earlier results between two unknown variables (Mahaboob et al., 2019). The logistics regression will produce a higher degree of relationship extraction between more datasets (Satish Kumar et al., 2019).