Under-Resourced Machine Learning for Stock Market Prediction

Abstract

In recent years, machine learning has made remarkable achievements in many fields. The stock market forecast is a vital application scenario. However, stock market prediction in real situations usually has relatively little data available for machine learning model training. This paper focuses on establishing a stock prediction model based on a small data set. The experiment mainly focuses on three aspects: data set, regression algorithm, and feature selection. To solve the small data set problem, we chose interpolation to expand the data number, improving the model's accuracy remarkably. After comparing the different combinations, such as any set of 2 features or any set of 3 features, this model finally used all 15 features to predict. Its prediction result is very close to the accurate stock price. In this model, MSE (Mean Squad Error) is a suitable option to evaluate the model's accuracy. The result shows the four regressions: Linear Regression, Polynomial Regression, Decision Tree Regression, Random Forest Regression. The result shows that Linear Regression has better-predicted accuracy. Based on the small data set, the best solution is to use as many features as possible to train Linear Regression.