Capital Markets Prediction: Multi-Faceted Sentiment Analysis using Supervised Machine Learning

Abstract

Over the years the stock market has proved to be very difficult to predict due to its unpredictable activities. Data mining techniques such as clustering, decision trees, genetic algorithms and artificial neural networks have been used in order to predict the stock market. Although there has been a significant amount of research done in this area, there are still many issues that have not been explored yet. The impact of fundamental analysis in the prediction of the stock market has been ignored though it can play a vital role in the prediction of the stock market. In this research, the problem of how a social data sentiment correlates to stock price is studied. A stock price prediction model was built using social data sentiments to predict the stock market. Sentiments analysis principles were applied to machine learning techniques in order to find the correlation between the stock market and public sentiments. This study particularly intended to assess the predictability of prices on the Botswana Stock Exchange through the application of Facebook sentiments classification. Three classification models were created that depicted news polarity as happy, calm, alert and vital. Results show that Naïve Bayes and Support vector machine performed well in both types of testing as compared to Random Forest. Naïve Bayes gave good results in terms of error margins with an accuracy of 83.3% making it the best classifier for our data set. When plotting the time series plot of sentiment scores and comparing it to the actual stock price graph, a conclusion can be reached that sentiments and stock prices are related and thus stock prices can be predicted using sentiments.