Towards Regression Model Selection Framework for Stock Price Forecasting

This paper presents a regression model selection framework based on the metric analysis for stock forecasting. Four machine learning algorithms are selected for demonstration: Multiple Linear Regression, Ridge Regression, K-Nearest Neighbour Regression, and Neural Network Regression, a type of deep learning regression. The mean square error is selected as the measurement metric for the forecasting models. Two stock datasets, Microsoft and Google, are selected for the simulation analysis. The simulation results show that no single method can outperform the others due to the diversity and uncertainty of the patterns of datasets. The framework may be able to recommend the better decision for a particular stock based on the forecasting metric.