Machine Learning Based Stock Price Prediction Using Sentiment Analysis from News Articles

In today's world predicting stock prices remains a challenge due to markets being volatile as it is driven by multiple factors. In the past, investors and business men depended on traditional methods of prediction, only taking pre-historic data for consideration. With the addition of introducing new factors that influence today's market and one that is highly overlooked is the market sentiment. Public perception always played a huge role in influencing the market and is crucial for investors to understand the height of the situation. Thus, exploring new areas that were previously neglected is quite the challenge. By introducing algorithms that gain a sentiment understanding from public media combined with various learning algorithm has proved the effectiveness of external factors that playa role in stock price fluctuations. Using a combination of algorithms such as FinBERT for sentimental analysis paired with Long Short-Term Memory (LSTM) has scored a predictive accuracy of 95%. Other algorithms that have proved notable performance are XGBoost, Gated Recurring Unit (GRU) and DeepAR.