An Enhanced Adaptative System based on Machine Learning for Predicting the Evolution of Islamic Stock Prices

Abstract

This paper suggests an enhanced machine-learning-based system to guide future stock price decisions. In reality, most existing machine learning systems, such as SEA (Stream Ensemble Algorithm), VFDT (Very Fast Decision Tree ), and online bagging and boosting, keep models updated with only new data and reduce training timeframes to allow working rapidly with the most recent model. However, limited learning times and the exclusion of essential information from previous data may result in a bad performance. When it comes to learning models, our system takes a different approach. It builds several models based on random selections of historical data from the main stock as well as related stocks. The best models are then combined to generate a final, performant model. We performed an empirical study on five Islamic stock market indices. We can say from the results that our system outperforms the existing published algorithms. This framework can contribute then to having an enhanced system that will enable different stakeholders to make rapid decisions based on the forecasted trend of indices.