Date | Tournament | Surface | Draw | Pls. | P. Pts. | P. Pct. | 1-st Favorite | 2-nd Favorite | 3-rd Favorite | Forecast |
---|

** Tournament Event Forecasts are based on the Tournament Simulation driven by Neural Network Match Prediction Algorithm*

Tournament Simulation is driven by individual Match Prediction. In each round, probabilities for each match in the draw are calculated using Neural Network Match Prediction Algorithm.

Lets name match probability that player A wins over player B as**Pm**_{A vs B}.

These probabilities determine the probability for each player to pass to the second round**Pr**_{A}(R2) = Pm_{A vs B}.

Probability of the player A to reach the next round R+1 is calculated this way:

**Pr**_{A}(R+1) = Pr_{A}(R) * Σ_{N=1-n} ( Pr_{N}(R) * Pm_{A vs N} )

This means that probability for player A to reach the next round R+1 depend on probability for player A to reach the previous round R multiplied by the weighted sum of probabilities for player A to win over his potential opponents in the next round. Weights of the potential opponents are the probabilities of each of the opponent to reach the round R.

That means for example, that probability for player A to win the title depends on the probability of the player A to reach the final as well as probabilities of all players in the other half of the draw to reach the final multiplied by probabilities for player A to win the final match over the each of them.

As tournament progresses, outcome of some matches gets known, thus the match probabilities are set to 100% and 0% for the winner and for the loser respectively.

Sometimes, as initial tournament draws are out, they include unknown qualifiers. Probability for the player to win over the unknown qualifier is determined by variation of the Match Prediction algorithm that includes player's winning percentage vs qualifiers, overall and by surface, level...

Lets name match probability that player A wins over player B as

These probabilities determine the probability for each player to pass to the second round

Probability of the player A to reach the next round R+1 is calculated this way:

This means that probability for player A to reach the next round R+1 depend on probability for player A to reach the previous round R multiplied by the weighted sum of probabilities for player A to win over his potential opponents in the next round. Weights of the potential opponents are the probabilities of each of the opponent to reach the round R.

That means for example, that probability for player A to win the title depends on the probability of the player A to reach the final as well as probabilities of all players in the other half of the draw to reach the final multiplied by probabilities for player A to win the final match over the each of them.

As tournament progresses, outcome of some matches gets known, thus the match probabilities are set to 100% and 0% for the winner and for the loser respectively.

Sometimes, as initial tournament draws are out, they include unknown qualifiers. Probability for the player to win over the unknown qualifier is determined by variation of the Match Prediction algorithm that includes player's winning percentage vs qualifiers, overall and by surface, level...

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 H2H percentages, recent form and vs hand (vs left-hander or vs right-hander).

Elo Rating neurons individually give high prediction rates, but when they are combined with H2H, recent form in short period and various winning percentages, the prediction accuracy is even further increased.

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 H2H percentages, recent form and vs hand (vs left-hander or vs right-hander).

Elo Rating neurons individually give high prediction rates, but when they are combined with H2H, recent form in short period and various winning percentages, the prediction accuracy is even further increased.