Exploring Property Price Prediction via Machine Learning and Deep Learning Approaches: The case of Pakistan

Abstract

In many real-world applications, predicting a property price is more realistic and appealing. Property price prediction helps individuals make better decisions when buying or selling a property. However, a large number of realistic tests must be conducted to determine the best methodologies and algorithms to find the optimal combination of these strategies for a reliable property price prediction model. In this research, we created a dataset of around 50k properties of Pakistan. This dataset has been used to train machine learning and deep learning models. In addition, we have applied feature selection methods to identify which attributes of a property can help to predict the best results in terms of evaluation metrics e.g., MAPE, RMSE, MAE, R2 , and MSE. Furthermore, after applying feature selection approaches such as the Pearson correlation coefficient, Mutual Information, Chi-Square, and the VIF on our dataset, we selected those characteristics that improved the reliability of our model. Then we applied ML and DL algorithms and evaluated them using various performance metrics. Afterward, the results revealed that the Extra Trees Reg. performed well as compared to other models of ML and DL. Moreover, after observing the outperforming results of ML & DL models, this research work finds that the data from Pakistan's real estate e-markets can be used for predicting property rates