Makine öğrenmesi modellerini kullanarak hisse senedi öngörüsü

Forecasting on stock market data is a challenging task due to dynamic and complex nature of the data in this field. Here, we present a comprehensive analysis on effects of using different data sources and deep learning architectures for forecasting stock prices. One of the biggest factors in the performance of deep learning models is effective representation of data. In this study, sentiment analysis was conducted on news articles that may cause financial external effects on stock markets, and various technical indicators extracted from the stock price datasets. Other than data representation, the performance of deep learning algorithms is dependent on hyperparameters, thus, we optimized various hyperparameters of the developed deep learning models using automated Bayesian optimization. We developed optimal Convolutional Neural Network (CNN), Long Short-Term Memory (LSTM), and Stacked Autoencoder (SAE) models for forecasting daily close prices from ten different historical stock price datasets traded on NASDAQ and NYSE target markets. Our study covers the effects of sentiment analysis on stock market forecasting and assessment of the proposed deep learning models in complex data representation. The evaluation of forecasting is analyzed with several error metrics. The obtained results revealed that the developed CNN model provides better performance than other ones. The integration of sentiment analysis and technical indicators improved the prediction performances, especially in the LSTM model. Forecasting on stock market data could be more robust when higher frequency data is coupled with automated feature extraction.