Neural Network Match Prediction Algorithm
Match Prediction is based on the Neural Network Algorithm with ~50 neurons for different features about players like Elo Rating, Surface Elo Rating, ATP Points,
Head-to-Head ratios and Winning Percentages varied by surface, tournament level, tournament, round, recency, match or set ratios, vs rank, vs hand, vs backhand...
Neural Network is trained on historical data for highest prediction rates to determine optimal feature weights.
During training, some neurons are determined to be useless and they are removed from the network, thus about ~30 neurons remain.
Elo Ratings, overall and by surface, are main contributors to the match prediction, following by recent form, H2H percentages and winning percentages vs hand (vs left-hander or vs right-hander).
Elo Rating neurons individually give high prediction rates, but when they are combined with recent form, H2H and various winning percentages, the prediction accuracy is even further increased.