Chess Position Advantage Prediction Using CNN and Image Processing

Summary

• Developed a CNN-based model achieving 69.43% hidden test accuracy and 68.76% validation accuracy for predicting chess position advantages (Black, White, Equal).
• Processed 22,000+ labeled chessboard positions, incorporating image features (SSIM, Canny Edge Detector) and chess-specific metrics such as center control, piece safety, clustering, mobility, pawn structure, and king safety.
• Optimized model performance through hyperparameter tuning (32-256 filters, dropout rates 0.2-0.3, learning rate 0.001).
• Implemented feature extraction and deep learning workflows using TensorFlow and Python.

Technologies Used

• Programming Languages: Python
• Deep Learning Frameworks: TensorFlow, Keras
• Image Processing Tools: SSIM, Canny Edge Detector, OpenCV
• Libraries: NumPy, Pandas, Scikit-learn
• Optimization Techniques: Hyperparameter Tuning