Investigation on House Price Prediction with Various Gradient Descent Methods

Abstract

As a branch of artificial intelligence, machine learning has received a lot of attention and has become a popular research field. It connects the huge database obtained from daily life with data prediction, which is of great help in decision making. In this project, we apply linear regression models and use different gradient descent methods, including general, mini-batch, and stochastic gradient descent, to optimize the error. We used convergence rate and accuracy as evaluation criteria, which will be explained via run time of convergence and mean square error respectively. For the regression models, both single-feature and multi-feature hypothesis models are investigated. In consideration of the risks of multicollinearity and overfitting, only simple linear model and its exponential transformation are adopted. The result indicates that mini-batch gradient descent has the biggest comprehensive advantage.