Forecasting Small Data Set Using Hybrid Cooperative Feature Selection

The aim of this paper is to propose the cooperative feature selection (CFS) to automatically select the critical factors that affect the performance of the forecasting performance of a small time series data. CFS sequentially combines grey relational analysis (GRA) and artificial neural network (ANN), which represents wrapper and filter method respectively. To test the efficiency of the proposed feature selection, it is employed to predict the total earnings of Malaysia Natural rubber based products. Results from the study shows that the proposed cooperative feature selections can increase the accuracy performance and learning time. Additionally, it also can work well in small data set and automatically choose the critical factor without human assistance.