Rough-Regression for Categorical Data Prediction based on Case Study

摘要

The conventional regression model is widely used to explain the mathematical relationship between exogenous and endogenous variables. This model is also capable for prediction and planning purposes. However, to achieve the high prediction accuracy is not easy task by using this model, especially for categorical data type, because fully uncertainty, volatility and unpredictable during data collection. Moreover, the data categorization and unclassified elements may significantly influence the prediction or classification accuracies. To handle both issues, the normality test for categorizing data and rough sets approximation for improving the performance of the conventional regression models are considered. Some date sets are examined to evaluate the both idea. The result showed that number of category for each attribute depended on its normality testing. Additionally, the data reduction was able to improve the accuracy of rough-regression model significantly.