Generating the Employment Predictive Model Using Data Mining Techniques

Abstract

This research demonstrates the generation of an employment predictive model using data mining techniques. The data set used in this research consists of employment statistics collected from Bank of Thailand (BOT), and 96 types of product exports data gathered from Thai Customs Department (TCD). Data are implemented by comparing three methods of Feature Scaling, which are Normalization, Standardization, and Non-Scaling, hybridized with three methods of Feature Selection, which are Correlation, Backward Elimination, and Non-Elimination. Afterwards, the predictive models are formulated by applying four data mining techniques: Linear Regression, Bayesian Ridge, SVR and XGBRegressor. In order to evaluate the efficiency of each predictive model, each model is measured by Mean Absolute Percentage Error (MAPE), Correlation and processing time. The experiment results indicated that the best-suited predictive model for the selected data was the Feature Scaling, achieved by Standardization combining with Feature Selection, executed by Backward Elimination, and Linear Regression technique. The Mean Absolute Percentage Error (MAPE) score was 1.247. Meanwhile, the correlation coefficient was 0.454 and the processing time took 2.125 milliseconds.