IPL2024: A review of MatchMind's H2H product

As the curtain falls on the IPL 2024 season, MatchMind is excited to share a comprehensive review of our performance, key learnings, and plans for the future. This season has been a landmark for us, showcasing the robustness and adaptability of our algorithms.

Algorithm Performance

Our models demonstrated impressive accuracy and return on investment (ROI) across different risk profiles:

• Model Accuracy: 71%

• Betting Accuracy: 70%

• ROI:

• Moderate Risk: 500%

• Conservative Risk: 290%

• Aggressive Risk: 600%
  

Live Modelling: The Game Changer

One of the standout aspects of this season was the effectiveness of our live modelling approach. By updating our models after each game, we mitigated the "primacy bias" that often hampers long-term predictions. This real-time learning allowed our models to adapt to the season's "key metrics," significance parameters, and weights dynamically.

Match Anomalies: The GT Mystery

A deep dive into the data revealed intriguing patterns, particularly in games involving the Gujarat Titans (GT). These anomalies have sparked curiosity and speculation. Here are a few notable examples:

• Match 17 vs. PBKS: Despite GT scoring 199, PBKS needed 41 off 12 balls. Unusually, three unexpected extras occurred in the 18th over, and Darshan, who hadn't bowled all game, was brought in for the 19th over.

• Match 32 vs. DC: GT was bowled out for just 89 runs.

• Match 40 vs. DC: Despite scoring 224, questionable tactics were observed, such as their top bowler, Sandeep Warrier, bowling only three overs, and Sai Sudharsan, who hadn't bowled at all, being pummeled for 22 runs in the 19th over.
  

Broader Application: BBL and NZ Super Smash

While we expect our models to perform well in the BBL and NZ Super Smash, the ROI may not be as high as in the IPL. The IPL's inherent uncertainty and close games create opportunities for favorable betting odds, which are less prevalent in other leagues.

Confidence Intervals: Reducing Risk

To manage risk better, we introduced "confidence intervals" in our predictions. If the prediction interval contained 50%, we advised against betting. Approximately 15 matches were deemed too close to call, reducing our exposure to high-risk bets.

Areas for Improvement

We have identified areas for further refinement and will be making adjustments during the off-season. Our goal is to have everything optimized for the Caribbean Premier League (CPL).

CPL and Pre-Determined Matches

Research suggests that some CPL games might be pre-determined, which could impact our performance. We are factoring this into our model adjustments and will proceed with caution.

New Features and Models

• 1st Innings Expected Runs Model: We will introduce a beta model predicting the number of runs scored by the first innings batting team, similar to an NBA "spread."

• Player-Level Model for BBL: This model will consider injuries, rosters, and player contributions to build accurate team ratings.

• Jackpot Function: When all 20 of our models predict a probability greater than 70% or less than 30%, we found a 78% accuracy rate. This "jackpot function" will be used to make confident predictions.
  

Our Models' Performance

Among our five core models, ML Mind 4 and 5 have been the most accurate at 73%, followed by ML Mind 1 and 3 at 70%. We only consider games for which we have provided bets, excluding 50-50 matches.

Importance of Confidence

The amount of money to bet is based on the model's confidence. For instance, a model predicting a 75% chance of winning suggests betting more than one predicting a 56% chance. This differentiation is crucial for maximizing winnings and minimizing losses.

Invitation to Betting Enthusiasts

We are opening up access to our Beta platform for the CPL, with caution regarding pre-determined matches, and will launch a fully functional MVP for the BBL and NZ Super Smash later this year. Get in touch if you're interested!

Season ROI and Strategic Insights

Our season ROI for a moderate strategy peaked around matches 45-55. The final leg of the season is highly uncertain, with teams vying for playoff spots. In the future, we may recommend pulling out around match 40-45, when games are treated more like round robins.

Stay tuned as we continue to refine and enhance our models, ensuring we provide the best possible insights and betting advice for cricket enthusiasts worldwide.