An Evolutionary Hyperparameter-Optimized Off-Policy Feature Selecion for Real Estate Tax Prediction
With: Pouria Zarei Hemmat, Javad Valipour, Roohallah Alizadehsani, PhD, Pawak Plawiak PhD
Precise predictions of real estate taxes (RETs) are critical for property owners and government agencies, influencing economic planning and public revenue generation. Traditional deep learning methods for RET often face difficulties in feature selection and are acutely affected by hyperparameter settings. To overcome these issues, we present a sophisticated predictive model for RET that incorporates a refined reinforcement learning (RL) technique, termed off-policy proximal policy optimization (Off-policy PPO), which excels in feature selection, and an advanced differential evolution (DE) algorithm for optimal hyperparameter tuning. The Off-policy PPO component constantly updates its feature selection based on data interactions, which sharpens its focus on crucial features and prevents overfitting. Moreover, the DE algorithm has been upgraded with a creative mutation strategy that utilizes k-means clustering to pinpoint crucial clusters effectively. We tested our model on diverse real estate datasets obtained from Kaggle, covering locations such as Bucharest in Romania, California in the USA, Helsinki in Finland, King County in the USA, and the Kingdom of Saudi Arabia (KSA). Our findings show that our model significantly outperforms traditional pricing tools, with losses ranging from 0.202 to 1.362 across various datasets. This superior performance underscores the capacity of the model for broad application in various markets, markedly improving financial analysis and strategic planning.