House Price Prediction With Statistical Analysis in Support Vector Machine Learning for Regression Estimation
Cesar Vasquez, Vinodh Chellamuthu · Curiosity Interdisciplinary Journal of Research and Innovation · 2021
Support vector learning machines are supervised models that analyze data for classification analysis. However, with modifications these models can be extended for regression estimation. In this paper, we use Dean De Cock’s Iowa house price dataset, composed of quantitative and qualitative variables, to build a predictive house pricing model. We employ a systematic approach to preparing the data and processing the features using statistical analysis. To reduce model over-fitting, we use stratified k-fold cross-validation. We also use various metrics for measuring model performance including r-squared (R2) and mean absolute error (MAE). Moreover, we explore various kernels for hyper-dimensional feature mapping to account for data that is not linearly separable. We compare the kernel performance and find that there is a trade-off between model complexity and performance. Furthermore, there also exists a trade-off between maximum accuracy achievable and general model performance.