Indonesia composite index prediction using Fuzzy Support Vector Regression with fisher score feature selection

Abstract

A precise forecast of stock price indexes may return profit for investors. According to CNNMoney, in the same month as much as 93% of global investors have lost money for trading stock. It is influenced by the stock market itself which is complex, nonlinear, has great noise, and chaotic system. Stock has characteristics that the higher the targeted potential return, the higher possibility of losses. One of the stock price indexes is stock composite index. Exact predictions of stock composite index can be critical for creating powerful market exchanging strategies. There are a lot of research that has been done to assist investors in minimizing possible losses. One of them is by predicting stock composite index. In this paper, a modified supervised learning method used to solve regression problems, Fuzzy Support Vector Regression (FSVR) is focused. As the complexity of many factors influence the movement of stock price prediction, the prediction results of Support Vector Regression (SVR) can not always meet people with precision. Thus, this study implies Fuzzy Support Vector Regression (FSVR) stock prediction model, in which fuzzy membership with mapping function is employed to generate a precise price fluctuation of stock. To assure the use of features on model prediction, Fisher Score is used to find high quality and informative features that can enhance the accuracy. Indonesia Composite Index or Jakarta Composite Index (JKSE) is considered as input data and the result showed that Fisher Score can be applied as feature selection on Indonesia Composite Index prediction with the best model is eleven out of fifteen features with 80% of training data with 0.043529 error.