The traditional metrics currently available for cricket enthusiasts and gamblers fail to provide deep insights into the performance of players. There are a number of statistical measurements used such as batting averages, strike rates, bowling economy rates and number of wickets taken. The ICC rankings also provide a useful guide of the ability of players and teams.
Whilst those metrics alone can give an indication of how good a player is they do not show us explicitly how much influence he has on the outcome of matches. For instance, two players could score a century in a test match but one scored coming in at 16 for 4 is much more valuable than one coming in at 401 for 4. But the record books would show and century and the average would be adjusted the same in both cases.
Similarly, a bowler who regularly dismisses two openers at the start of an innings is more valuable than one who routinely mops up the tail. This would not be shown in the regular statistics.
The project behind this website set out to develop a methodology that can measure the all bowling and batting performances with one comparable metric. Past performances are then combined to produce a batting and bowling master rating for every team and current player for the different formats of the game.
The ratings are a result of crunching thousands of past matches using a proprietary algorithm. The algorithm combines reinforcement learning and neural networks to self-learn the influence of performances on the final outcome of matches.
The card below shows the 1st South Africa innings in the 1st Test versus England in 2015. England had scored 303 in their first innings.
England's bowlers performed well and dismissed South Africa for 214. The algorithm allocated Broad 797 and Ali a considerably less 319 despite getting the same number of wickets. Finn got a negative rating of -102 despite getting 2 wickets. Why does the algorithm do this?
Looking at the scorecard more in depth we can see Broad dismissed 4 of the top 6 batsmen whereas Ali's wickets were in the middle and lower order. Broad also had a better economy rate. Those two factors combined resulted in a higher rating. Finn's wickets were 10 and 11 in the batting so he did not receive very much credit from the algorithm for them.
Just looking at the bowling card or player summary statistics it would be hard to work out how much better Broad's performance was than Ali's. The algorithm gives a precise logical metric for this.
The World T20 final in 2016 was one of the most dramatic cricket matches of all time. The batting card below shows the algorithm allocated Braithwaite 916 points. He scored 34 runs off 10 balls at a strike rate off 340 runs per hundred balls. In comparison Marlon Samuels was given just 91 points for scoring 85 runs off 66 balls.
The records for future generations will show Samuels top scored in the final and his average will gain far more than Braithwaite's as a result of the match. So why did the algorithm allocate Braithwaite so many more points than Samuels?
The answer is the scorecard alone does not explain the full drama of the match. It is true to say that Samuels held the innings together as wickets fell around him. But he failed to increase his strike rate as the run rate required increased. This left a seemingly impossible 19 required to win off the final over.
On the Betfair betting exchange England had an implied probability of winning of around 90% at this point in the match. Carlos Braithwaite then proceeded to smash Ben Stokes for four sixes in a row to win the trophy for the West Indies with two balls to spare. Braithwaite had the biggest impact on the result of the match given the situation he found himself in. This is the reason the algorithm has given him such a high match batting rating.
Every batting and bowling performance is given a rating for every match. The Master Ratings are combined metrics of all previous matches. Each player that has faced or bowled at least 200 deliveries in a format within the last year is given a Master batting and/or bowling rating for that format.
The Master ratings are calculated using neural networks to predict how a player will perform in a future match.
The table below shows the top 5 test batsmen using Cricsq and their corresponding ICC ratings (26th March 2017). The artificial intelligence based Cricsq ratings and the traditional ICC rankings have 3 of the same top 5 batsmen. This shows that algorithm has taught itself a reasonable methodology.
|Name||Cricsq Rank||Cricsq Rating||ICC Rank||ICC Rating|
Every Test, One-Day International, T20 International and Major T20 League team is given a rating. These ratings are calculated using the last 3 years results for international matches and the last year for T20 League teams. This can be a problem early season in T20 Leagues if there is a big change in personnel. In the pre-match statistics the ratings of current squad is always shown.
The table below shows the top 5 ODI teams using Cricsq and their corresponding ICC ratings (26th March 2017). The results are the same except that the Cricsq ratings have a wider spread than the ICC ones. There are also ratings provided for international teams using just one year's data. This can help cricket fans and punters see if their team is in form.
|Name||Cricsq Rank||Cricsq Rating||ICC Rank||ICC Rating|
Cricket Betting Bookmakers
India beat Pakistan (8 wks : 126 balls left)
India beat Hong Kong (26 runs)
Afghanistan beat Sri Lanka (91 runs)
Trinbago Knight Riders beat Guyana Amazon Warriors (8 wks : 15 balls left)
Pakistan beat Hong Kong (8 wks : 158 balls left)