Critical Analysis of History Bits Algorithm with Attribute Ranking Strategies for Stock Market Prediction Accuracy Evaluation

Prediction of stock market trading decision is a classical problem. Different machine learning models can be used to solve this prediction problem and help traders to get best return on investment. In an attempt to solve this prediction problem, we have developed a novel History Bits based algorithm. This algorithm accepts 75 technical indicators prioritized using attribute ranking strategies and transformed into buy and sell trading signals. All 75 indicators are partitioned into certain number of groups based on their rankings and bias is added as per their groupings. There are many attribute ranking strategies like Pearson correlation coefficient, information gain, gain ratio, OneR, relief evaluator and, symmetrical uncertainty evaluator. Different ranking strategies may produce different importance levels of attributes impacting the trader’s decision. In order to overcome this problem we propose an ensemble based attribute ranking strategy to collect combinatorial effect of all attribute ranking strategies. For the experimentation, a real dataset of NIFTY 50 index for a period of 20 years is used. Prediction accuracy of History Bits algorithm using different ranking strategies is compared. The proposed algorithm is quite stable with prediction performance over different attribute ranking strategies.