Property Rate Forecast Using Machine Learning
Abstract
In our research, we are predicting the selling prices of dwellings in King County, Washington DC by applying various Machine Learning algorithms for improving dwelling market interpretation and appraisal of differentiating among market value and market price. The dataset considered in our case study contains dwelling data of 21613 properties as well as contains 20 analytical features. In past various linear based and non-parametric algorithms are used for forecasting real estate price but are ineffective in observing anomalies in the data. Therefore, we used ensemble-based algorithm like Random Forest and deep learning-based algorithm like Neural Networks to perform a comparative analysis against the typical used algorithms like Multiple Linear Regression, Polynomial Regression and K Nearest Neighbors. We further performed Deep Data Exploration using Histogram, Pair Plot Visualization and Correlation among features during Data Analysis for understanding behavioral nature of the data. Comparative analysis was performed based on various performance metrics such as k-Fold Cross Validation, Adjusted R-Square, R-Square and Root Mean Square Error. According to the performance metrics considered for dwelling price prediction, Random Forest outperforms compared to other models and can be further used in order to help buyers and sellers make better decisions based on dwelling price valuation.