Machine Learning Based Approaches for Predicting Stock Closing Price

Due to random, uncertain and dynamic nature of stock market, predicting the price is quite a tough and challenging task. Dominance of machine learning in all sectors of life is noticeable. Hence, researchers have paid attention on predicting stock price through different machine learning architectures. This study has focused on observing performance of six machine learning techniques and has proposed three hybrid machine learning approaches which consists of Light Gradient Boosting Machine for feature selection and Recurrent Recurrent Neural Network, Long Short Term Memory and Gated Recurrent Network for prediction. The hybrid approaches showed better results. But among all of them, hybrid model of Light Gradient Boosting Machine and Gated Recurrent Network performed best. Data of seven years is collected from two technology groups and is pre-processed. Eight technical indicators are utilized as inputs which are obtained from existing values of dataset. Performance comparison is observed through different evaluation metrics: MAE, MSE, RMSE and R
^{I. 2}
. The investigational output proves the dominance of the presented approach over the other implemented models.