• ANIL KUMAR DUBEY
• Karan Singh Rawat
• Md Mashhood Raza Siddiqui

Game playing in chess is one of the important areas of machine learning research. Though creating a powerful chess engine that can play at a superhuman level is not the hardest problem anymore with the advent of powerful chess engines like Stockfish, Alpha Zero, Leela chess zero etc. Most of these engines still depend upon powerful and highly optimized look-ahead algorithms. CNN which is used primarily for images and matrix-like data is been proved successful with games like chess and go. Treating chess like a regression problem. In this paper, a supervised learning approach is proposed using the convolutional neural network with a limited look ahead. Data was collected around 44029 chess games from the FICS chess database with players having an Elo rating of 2000 and above. Our goal is to create a zero-knowledge chess engine. The trained model is then paired with a minimax algorithm to create the AI. Our proposed supervised system can learn the chess rules by itself from the data. It was able to win 10% of the games and draw 30% of games when manually tested against Stockfish computer engine with Elo of 1300. CNN can detect various tactical pattern to excel in games like chess even when using a limited lookahead search.

CNN, Backtracking, look ahead algorithm, Evaluation Function

"AI BASED CHESS ENGINE", International Journal of Emerging Technologies and Innovative Research (www.jetir.org), ISSN:2349-5162, Vol.9, Issue 5, page no.g742-g744, May-2022, Available :http://www.jetir.org/papers/JETIR2205787.pdf

"AI BASED CHESS ENGINE", International Journal of Emerging Technologies and Innovative Research (www.jetir.org | UGC and issn Approved), ISSN:2349-5162, Vol.9, Issue 5, page no. ppg742-g744, May-2022, Available at : http://www.jetir.org/papers/JETIR2205787.pdf