Machine Learning Systems in Sports


I’m often asked about how to make accurate predictions and what tools I used to employ in the past. I’ve observed that this subject is not well understood in the NBA community, so I’ll try to provide more insight in this post.

The Majority of people picked the Lakers against Dallas, right? Simulations, statistics, and gut could tell us the Lakers were the better team, as all players were healthy and they’ve also been supposed to have the motivation of pursuing the three-peat, etc. They ended up getting swept by Dallas and arguably the best coach in NBA history admitted after the series that despite the way how it ended, he was happy to see things have come to an end.

Many individuals and organizations have a shot at predictive modeling. The art of predicting anything brings numerous techniques to the table. Simulations, the well-known way of predicting, can reveal hidden patterns or potential successes/failures about any team or an individual player. However, trades and injuries are the factors that ruin simulations and have a big impact on prediction accuracy.

Like what simulations do, machine learning techniques can help hidden data trends come to surface. Aside from statistical prediction, machine learning techniques are another method of providing sport-related predictions.

Neural Networks are considered to be one of the strongest machine learning systems in sports prediction. Within neural networks, data sets are learned by the system, and hidden trends in the data can be unveiled.
Neural Networks can be used as a decision support system and as a fantasy league tool in the following ways;
+ Stat Mining: Location of common characteristics in large amounts of data,
+ Odds Setting: Calculation of winning probabilities in a sports game,
+ Momentum Mentality: Forecasting trends based on previous data,
+ Similar Players: Grouping potential assets/talents based on similarities.

Other machine learning techniques include genetic algorithm, the decision tree algorithm, and a regression-based variant of the Support Vector Machine (SVM) classifier, called Support Vector Regression (SVR).

The NBA has been experimental for various machine learners so far. Despite employing different algorithms, they all had one common similarity, beating the NBA odds for a competitive or financial advantage. For example, according to the hot-hand theory in basketball, researchers found that making a shot would not increase the chance of making another shot.

The academical study based on real data showed that the success of a shot was independent of previous shots.
What’s more, people frequently try to find out a pattern that never exists or even happened in the past. Streaks are possible and can be observed frequently, but future games must be evaluated independently from the result of previous games. This is due to the nature of being normally distributed. Never forget that many games come down to critical free throws the players even who have the best shooting mechanisms might miss free throws in the last second. This situation triggers other game result-oriented events and has impact game-winning spreads or game totals. At this point, psychology science would better take things over. Simulation and predictive techniques are the tools to identify weaknesses if correctly used the best action for fixing errors. These techniques begin to become arsenals to isolate instances of these human factors.

The ability to apply statistics and rigorous mathematical models to provide instantaneous results is still at a very early age, observing what they will develop into in the near future would be interesting.