Dwelling Price Prediction Based on Physical, Economical & Social Indicators Using Regression Methods
The domain of dwelling prediction is historically done by exploiting typical linear regression methods; however the relationship among price and property features exhibits complexity as well as non-linearity, which are not handled robustly by theexistingresearch. Therefore, in our study, variety of advance regression techniques are introduced, including a series of ensemble models which have the capability to learn non-linear data, handle outliers and can strongly predict residential prices. In thepaper,we considered an open-sourcedata recordthat includes 20 descriptiveattributes and 21,613 records of propertytransactions in King County, USA from May 2014 to May 2015. Data Pre-processing, feature engineering, data normalisation, standardisation and exploratory data analysisare conducted on our data to form a structured, correlative and complete dataset to beusedin trainingand testing. Further, the tuning of ensemble modelsis enforced byexploiting hyperparameters to boost the accuracy. Finally, ourexperimentsconclude that the Cat-Boostalgorithmresulted in an accuracy of 91.45 %that outperforms other models by a close proximity while doing manual tuning of parameters whereas XG-Boost imparts the most optimized response having an accuracy of 90.18% while performing automated tuning of parameters.