Stock market turning points rule-based prediction / Lersak Photong … [et al.]

Stock market turning points can benefit stock market investors when making decisions during stock market trading. The main objective of this study is to investigate the effects of online news towards stock market turning points. This investigation involves three aspects: studying the methods of news sentiment analysis and rule-based optimisation, analysing the data and comparing the performance of models in order to obtain the most accurate prediction to provide recommendations on how to obtain the most accurate predictions for stock market turning points. Seventeen companies’ data were taken from the Yahoo! Finance website. Feature extraction was used for classifying relevant vocabulary into the same category of macroeconomic factors. Feature selection was used to sort out key features for further classification. News classification into factors affecting stock market turning point was done using Naïve Bayes, Deep Learning, Generalized Linear Model (GLM) and Support Vector Machine (SVM). Simultaneously, news sentiment analysis techniques were used to discover the polarity of news according to each factor. From news classification and news sentiment, a rule-based algorithm was used to predict the stock market turning points. Finally, rule-based optimisation techniques such as Particle Swarm Optimization (PSO), Differential Evolution (DE) and Grey Wolf Optimizer (GWO) were used to minimise the amount of time employed in the stock market turning points prediction. Results show that the best feature selection is term frequency and trimming of the feature with a frequency greater than 95%. The best news classification approach is based on Deep Learning techniques that provide the most accurate classification. The study suggests that the application of rule-based optimisation to predict stock market turning points generate more accurate and time saving decision.