Optimization of linear regression in house price prediction

Abstract

House price prediction plays a very important role in housing transactions. Linear regression based algorithms show good effects in predicting house prices. They have strong interpretability and fast operation speed. However, people ignore the estimation of deviations in linear regression (LR) algorithms. In this paper, k-nearest neighbor (KNN) algorithm is supposed to estimate deviations that are added to the result of linear regression to predict house prices accurately. Furthermore, deviation regression (DR) algorithm is supposed to make the prediction result more accurate. By utilizing Boston House Price data from Kaggle, extensive experiments are conducted and demonstrate the superior performance and compatibility of DR.