Stock Market Forecasting Using SVM With Price and News Analysis

Sammanfattning: Many machine learning approaches have been usedfor financial forecasting to estimate stock trends in the future. Thefocus of this project is to implement a Support Vector Machinewith price and news analysis for companies within the technologysector as inputs to predict if the price of the stock is going torise or fall in the coming days and to observe the impact on theprediction accuracy by adding news to the technical analysis.The price analysis is compiled of 9 different financial indicatorsused to indicate changes in price, and the news analysis uses thebag-of-words method to rate headlines as positive or negative.There is a slight indication of the news improving the resultsif the validation data is randomly sampled the testing accuracyincreases. When testing on the last fifth of the data of eachcompany, there was only a small difference in the results whenadding news to the calculation and such no clear correlation canbe seen. The resulting program has a mean and median testingaccuracy over 50 % for almost all settings. Complications whenusing SVM for the purpose of price forecasting in the stockmarket is also discussed.