Comparison of Model Performance on Housing Business Using Linear Regression, Random Forest Regressor, SVR, and Neural Network

This study investigates the effectiveness of four machine learning models for predicting house prices: Linear Regression (LR), Artificial Neural Networks (ANNs), Random Forest Regressor, and Support Vector Regression (SVR). As the demand for housing continues to rise due to population growth, the housing business has become one with high long-term potential. This is evidenced by an upward trend in house prices in Indonesia and other regions. For instance, in the United States, new single-family home sales climbed 8.8% to 693,000 units (SAAR) in March 2024, marking the fastest pace since September 2023 and an 8.3% year-over-year increase. Property values are influenced by various internal factors, such as area and number of rooms, as well as external factors like inflation rates and government policies. Linear Regression (LR) is known for its simplicity and ease of interpretation, providing high accuracy with low computational cost. In contrast, Artificial Neural Networks (ANNs) excel at capturing non-linear relationships, making them suitable for more complex datasets. Random Forest Regressor, an ensemble learning method, enhances predictive performance by averaging multiple decision trees and controlling overfitting, while SVR is effective in high-dimensional spaces and can model non-linear relationships using kernel functions. We employ the Boston housing dataset, a widely used benchmark for house price prediction tasks, to compare these models. The dataset is split into 70% for training and 30% for testing. The performance of the models is evaluated using Mean Squared Error (MSE), R-squared, and Mean Absolute Error (MAE). The results show that Linear Regression achieves an MSE of 0.0106, R-squared of 0.67, and MAE of 0.075. The Artificial Neural Network achieves an MSE of 0.0046, R-squared of 0.86, and MAE of 0.047. The Random Forest Regressor achieves an MSE of 0.0060, R-squared of 0.81, and MAE of 0.050, while the SVR achieves an MSE of 0.0054, R-squared of 0.83, and MAE of 0.056. By comparing these models, we find that the Artificial Neural Network (ANN) achieves the highest accuracy, followed by SVR, Random Forest Regressor, and Linear Regression. These findings provide valuable insights for buyers, sellers, and developers in the housing market, highlighting the potential of advanced machine learning models in accurately predicting house prices.