Stock Price Prediction Using Machine Learning and Sentiment Analysis

The biggest challenge in computational finance is forecasting stock prices. The stock prices are always in a state of fluctuation due to various economic, geopolitical, and investor sentiment issues. None of the modeling techniques has been considered reliable in forecasting stock prices. The traditional models are based on past stock price movements; therefore, they are not able to consider major fluctuations in the stock market due to news and announcements. To overcome all these challenges, we are proposing our solution: "StockPredict AI." Our proposed system is a hybrid interactive application created using React. Our proposed system is "StockPrediction." Our system includes 38 major stocks in the NASDAQ stock exchange, including AAPL, NVDA, TSLA, AMZN, GOOGL, etc. In our proposed system, we are using two unique data sets: a quantitative data set and a qualitative data set. The quantitative data set includes financial ratios, stock performance, and stock price movements. The qualitative data set includes social media feed analysis. Historical prices and qualitative sentiment data from financial news headlines. Instead of using these data streams separately, we have proposed a unified approach to both data streams. For the data stream containing numerical data, we have proposed Long Short Term Memory (LSTM) networks with 50 units and 0.2 dropout. In addition to that, we have proposed Random the data stream containing sentiment signals, we have proposed a fine-tuned FinBERT model. The proposed FinBERT model can be used to classify eight different sentiment types: positive, negative, neutral, bullish, bearish, uncertain, fear, and greed. In addition to that, it can also be used to classify emotions based on Plutchik's wheel of emotions. Real-time macroeconomic data has been used in our proposed system. The data stream used in our system is VXN, USDX, and UNRATE. Our proposed system has achieved an accuracy of 80.5% along with an RMSE . This proposed system is an enhancement of 36.5% over the standalone LSTM system. It can be inferred that using a sequence of stock prices along with macroeconomic data and multi-class sentiment data can be used to achieve more accurate predictions