Improving machine learning prediction performance with explainable artificial intelligence in financial time series prediction

Recently, significant efforts have been made to successfully predict the direction of stock prices using machine learning techniques. Machine learning techniques and models are inherently referred to as black boxes in the literature, meaning that no inference can be made about how the model makes its predictions and inferences. Solution proposals for this issue are being developed under the umbrella of "explainable artificial intelligence". In this study, we suggest using an explainable AI approach that can be used to evaluate the reliability of the predictions of existing models instead of developing more successful machine learning techniques, thereby allowing decision-makers to avoid bad decisions that are responsible for the overall prediction performance decrease. Because if there was a measure of how certain the prediction model is about any given prediction, relatively more reliable predictions could be used for decision-making, and lower-quality decisions could be avoided. In this study, a new two-stage prediction model based on machine learning, explainable artificial intelligence and empirical mode decomposition is proposed for predicting the direction of the stock market. According to the findings of the study, the prediction model produces more successful results than existing prediction models in the literature.