Data Balancing Modification for Multinomial Logistic Regression Model

Big data and data analysis are very necessary and play important role in the operation of organizations especially the future forecasting. To improve the predictive accuracy, an imbalanced data problem was considered before establishing a proper multinomial logistic regression model. Dealing with such problem, Random over sampling, Random under sampling, and Hybrid approaches were considered. Moreover, Simple random sampling and Stratified sampling techniques were applied in the data modification stage. The results revealed that the models fitting information of both imbalanced and balanced data models were similar and the model accuracies were around 70%. Although the accuracies were not different, the predictive accuracy and the precision of the minor group were increased.