Neural network hyperparameter optimization for prediction of real estate prices in Helsinki

Area of Science:

• Real Estate Economics
• Machine Learning Applications
• Computational Finance

Background:

• Accurate real estate price evaluation is crucial for stakeholders including investors, banks, and property owners.
• Artificial Neural Networks (ANNs) show potential for real estate valuation but are sensitive to hyperparameter settings.
• Hyperparameter optimization is key to enhancing ANN performance in predictive modeling.

Purpose of the Study:

• To apply and optimize an Artificial Neural Network (ANN) model for precise real estate price prediction in Helsinki, Finland.
• To investigate the impact of hyperparameter fine-tuning on ANN model accuracy for property valuation.
• To leverage Bayesian optimization for efficient ANN hyperparameter tuning.

Main Methods:

• Implementation of an Artificial Neural Network (ANN) architecture for real estate price prediction.
• Utilizing Bayesian optimization algorithm to fine-tune ANN hyperparameters, including activation functions and optimization algorithms.
• Evaluation of model performance using standard metrics such as Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), and R-squared (R2).

Main Results:

• Model optimization through hyperparameter tuning led to improved performance across all evaluation metrics.
• The optimized ANN model achieved a relative mean error of 8.3% in real estate price prediction.
• Graphical illustrations complemented the quantitative metrics in assessing the model's predictive accuracy.

Conclusions:

• Hyperparameter optimization is essential for maximizing the effectiveness of ANNs in real estate price prediction.
• The Bayesian optimization approach proved effective in enhancing ANN model performance for property valuation.
• The study demonstrates a significant improvement in prediction accuracy, offering a more reliable tool for real estate market analysis.