Penalized ensemble feature selection methods for hidden associations in time series environments case study: equities companies in Saudi Stock Exchange Market

Abstract

The problem of optimal subset selection from a large number of time series is addressed in this work using machine learning of financial forecasting. This is a persistent problem in stock market that is largely due to the vast amount of daily-basis data points which require sensitive and robust intelligent data analysis techniques for capturing hidden associations between time series features. In this work, we are interested in generalizing the concept of capturing hidden associations between predictors from financial time series data points in the setting of penalized ensemble feature selection techniques. We have shown how recently developed penalized ensemble feature selection methods are capable of revealing hidden and informative dependencies between equity companies that appear in Saudi Stock Exchange Market in different daily time series datasets. The results have shown that our methods outperformed the well-known lasso regularization method particularly for small sample size.