Application of Machine Learning and Artificial Neural Networks to Predict Real Estate Sales

This paper aims at developing forecasting models to predict the sales of a real estate project in India using variables like amenities, neighbourhood characteristics, price and registration of real estate project. The predictive models include multiple linear regression (MLR)/ordinary least squares (OLS), random forest (RF) and artificial neural network (ANN) to discuss the efficacy and popularity of real estate forecasting models, their results and biases. Discussed models are divided into four categories: macro-economic forecasting models, micro-economic forecasting models, time-series models and spatial hedonic models. Hedonic regression models are one of the most common models adopted by researches followed by artificial neural network (ANN) and time-series models. Random forest and ANN models suggest that nonlinear algorithms show better performance in the prediction. Use of ANN outperforms the use of other methods such as the OLS. Prediction accuracy is highest in neural network, followed by random forest and lastly by OLS for the model tested. Finally, the inference from this study is that the forecasting models require recalibration over time and require variables updating for forecasting efficacy. Limitation of their use and benchmarking depends on country, location, culture and context, with limited portability.