ABSTRACT
Stock price prediction (SPP) plays a major role in financial marketing since it is difficult for sellers or buyers to forecast future stock values. The tendency of a stock price is very complicated as it depends on various factors. Much research has been done to forecast stock prices which is helpful for users to identify the direction of stock price movement. Recently, machine learning and bio-inspired algorithms have been employed for precise SPP. Though classification methods perform SPP effectively, the presence of numerous factors in the stock process decreases the efficiency of the applied classification algorithm. Feature selection methods can be used to decrease the computational complexity and enhance the classification accuracy of SPP. In this paper, two bio-inspired algorithms, namely genetic algorithm (GA) and particle swarm optimization (PSO), are used for feature selection. They eliminate the unwanted and irrelevant features and choose significant features to model the classification system. The goal of this work is to investigate the effect of feature selection approaches in the ant-miner-based classification task of SPP. The performance is validated by testing the proposed method against four datasets: the Dow Jones dataset and three datasets gathered from Yahoo! Finance on a daily, weekly and monthly basis. The empirical results state that the PSO-based feature selection for SPP using ant-miner algorithm performs well, and it is noted that the classification accuracy is increased by the inclusion of feature selection method in SPP.
