Feature Selection for financial data – comparison

Abstract

Data analysis is currently one the key for the success of good condition of the companies. Feature selection as a preprocessing of data method the estimator accuracy scores can be improved, as well the performance on very high-dimensional data set can be boosted.The paper presents an application of the filter Feature Selection method to check how the data set after the application of the selected method will look like. The goal of this article is to compare the feature selection method with the application without usage of Feature Selection about the set of prepared financial data. For this purpose, the experiments have been executed with usage of Jupyter Notebook with Python as a programming language. The data source for the data set was Stooq, which is an open source website for tracking of changes on the financial markets. For this purpose, several joint-stock companies of various sectors from Polish Stock Exchange have been selected. The Feature Selection has been applied before the data has been entered into the Artificial Neural Networks model. The results have been compared using the MSE value. The lowest MSE value has been obtained where no Feature Selection method has been applied, however MSE value for Filter Method is one order of magnitude smaller. For the experiments without Feature Selection 112 input variables have been used, for the Filter Method only 4 variables have been incorporated.