Stock Buy/Sell Prediction Based on Technical Analysis and News Sentiment Using Deep Learning

The stock market is a compelling arena for investors and researchers. Its movement is influenced by various global factors, including public information, social conditions, macroeconomic policies, and the political climate of a nation. Given the multitude of factors, news regarding stocks can serve as a significant criterion in the decision-making process for buying or selling stocks. This study employs pretrained transformer models, including FinBERT, FinancialBERT, and FinBERT-tone, to evaluate sentiment in stock news. Technical stock indicator features, including Moving Average, Moving Average Convergence Divergence (MACD), Bollinger Bands, and Relative Strength Index (RSI), are utilized to enhance the model. The two datasets are subsequently merged by modifying the index, and the training procedure is executed utilizing deep learning and ensemble learning techniques in the format of a Voting Classifier. These methods pertain to four categories of Recurrent Neural Network (RNN) architectures: Long Short-Term Memory (LSTM), Bidirectional Long Short-Term Memory (Bi-LSTM), Gated Recurrent Unit (GRU), and Bidirectional Gated Recurrent Unit (Bi-GRU). The Voting Classifier model employs comprehensive deep learning techniques, leveraging all technical indicator features and sentiment labels to achieve optimal accuracy when evaluated on BBRI stock, targeting a 50-day period, with four classification labels and a 30-day sliding window. The accuracy achieved with this strategy was 94.22%