A comparative Study to Predict the Property value using Machine Learning

Abstract

Estimating the value of a property in terms of money can be a very difficult challenge. A good estimation can help both buyer and seller and not also there is a huge demand for the models that can estimate the value of the property more precisely as it can be hugely helpful to avoid possible loss while trading in the property which is beneficial for both buyer and seller. Accordingly to solve this issue a lot of researchers have proposed a lot of Machine Learning and Deep Learning regression algorithms and models like Back Propagation Neural Network, Fuzzy Logic, Arima model, Multilevel Modelling, etc. Some of these models include some optimization or boosting techniques like Swarm optimization and Adaboost which help the model to give more precise results. Some of these previous models will be discussed further in this paper. To predict the property value with maximum effectiveness, we have conducted a comparative study of different Machine Learning Algorithms along with some attribute selection technique Partial Least Square Regression (PLSR), k-folds cross-validation, and pre-processing techniques to boost the accuracy of mentioned models. Hereby the performance will be evaluated on four parameters using the same dataset which will help us to compare the performance of each algorithm. These Four parameters are Average Profit or Loss, Adjusted R-Squared, Mean Absolute Error, and Mean Squared Error. Also, we have introduced a hybrid model to overcome the mentioned problem and this will be discussed further in this paper. Finally looking at the results obtained we can use the best algorithm to solve this problem. The algorithms used in this paper are Kernel Support Vector, XGBoost, and Decision Tree, ElasticNet, and a Hybrid regression model. According to the results obtained the Hybrid Regression model proposed by us is best for the estimation of property value.