Machine learning-based stock price prediction

In this study, research has been conducted on the price analyses of companies operating in the petroleum refinery sector in five different countries (Turkey, France, America, Austria, and England) and the significance levels of variables affecting the price. The aim of the research is to predict the stock price, which is the dependent variable, using various independent variables. The relationship between the dependent variable and a total of 74 independent variables in six categories has been examined. These categories include technical indicators, fundamental indicators, country indices, price-related variables, oil production data, oil prices, and exchange rates. Correlation analyses have been conducted between the variables examined with each stock. The Rapid Miner Auto Model module has been used for correlation analyses. Significant results have been found among the price-related variables for five different stocks. The variables with the lowest correlation values were technical indicators and oil production data. The strongest relationships were identified as fundamental indicators and country indices. The research aimed to answer the question of which variables are important for predicting stock prices. Predictions were made using the Rapid Miner program. The algorithms used in the study are Linear Regression, decision tree, deep learning, random forest, gradient boosting, KNN, artificial neural network, and SVM. Evaluation metrics such as RMSE, MAE, and MAPE were used to assess the accuracy of these predictions. The best results for Chevron, OMV, Total, and Tüpraş companies were obtained with the SVM method. For Shell, the best result was found with the artificial neural network method.