Stock price trend prediction method based on support vector machines with Fisher score

Abstract

Recently, stock investment has become very developed and carried out by people from various backgrounds, not only the successful entrepreneurs but also the younger generation. However, due to its complex, nonlinear, noisy, and chaotic system, stock has characteristics that the higher the targeted potential return, the higher possibility of losses that must be accepted. Predicting the movement of stock prices trend is one of ways to evaluate and anticipate future stock prices. This study proposes a method based on Support Vector Machines (SVM) using Fisher Score as the feature selection to predict stock price trend. Fisher Score is one of the widely used supervised feature selection methods. Fisher score chooses a suboptimal subset of features to identify the most influence one in prediction. While, Support Vector Machines (SVM) deals with nonlinearly separable input, gives impact to the accuracy. As Fisher Score is used as feature selection, on the preprocessing data, this study was using technical analysis with fifteen technical indicators that helped to create the possibly useful features. Daily stock prices of PT Bank Negara Indonesia (Persero) Tbk. (BBNI), PT Bank Rakyat Indonesia (Persero) Tbk. (BBRI), PT Bank Central Asia Tbk. (BBCA), and PT Bank Mandiri (Persero) Tbk. (BMRI) were considered for the analysis. The model evaluation with the proposed method showed that Support Vector Machines with Fisher Score yielded the best result, in regard to the accuracy. Among the four data sources, the best model fit was obtained for data on PT Bank Mandiri (Persero) Tbk. with eleven out of fifteen technical indicators, 70 % of training data, σ = 0.01, and 99.66 % accuracy.