Optimization of the Random Forest Method Using Principal Component Analysis to Predict House Prices

Abstract

Investment is an interesting thing, especially property investment. The developer must also be careful in determining the price of the property. It should be noted that every year, both short-term and long-term, property prices increase and rarely go down. In determining the price, it is often also based on the features of the house such as the concept, location, bedrooms, etc. To predict house prices based on their features, the random forest has a good performance for predicting house prices. However, the random forest method has the disadvantage that if you use too many variables, the training process will take longer and feature selection tends to select features that are not informative. One way to reduce features without removing other features is to use Principal Component Analysis. In this research, the method used is Principal Component Analysis (PCA) and Random Forest. From the results of model training, it can be concluded that the use of model evaluation results using PCA has a smaller error rate and more consistent values, with an average of 0.018. While the results of the evaluation without PCA and using only Random Forest have a higher error value with an average of 0.03125. The training time using the PCA model has a faster time, with an average of 7918 milliseconds, while those using only random forest without PCA have an average time of 8975 milliseconds.