The PredictPal Philosophy
PredictPal was born out of a common frustration in the sports analytics industry: the lack of transparency and the sensationalism in predictions. Most services claim unrealistic success rates without providing an auditable history or explaining the margin of error of their models.
We believe that statistical honesty is the only viable path. We don't hide our errors; we measure them. That is why we evaluate every conditional probability prediction we generate using the Brier Score, a standard metric in meteorology and probabilistic forecasting that measures the accuracy of assigned probabilities.
How Does Our Model Work?
Our algorithms analyze millions of historical data points from over 220,000 matches in the most active table tennis leagues (Setka Cup and TT Elite Series).
Instead of only estimating who will win a match before it begins (a basic "prior"), we calculate the conditional victory probability as the match evolves. Specifically, we model player behavior after the first two sets are decided, evaluating transitions under 4 critical states:
- 2-0: Home player leads two sets to zero.
- 1-1 (HA): Home player won Set 1, and Away player won Set 2.
- 1-1 (AH): Away player won Set 1, and Home player won Set 2.
- 0-2: Away player leads two sets to zero.
Each of these estimates is accompanied by a 95% Wilson Confidence Interval, allowing visual representation of the model's precision based on the volume of historical matches played between the two players or within that specific league.