Before March arrives, millions of people all over the world will fill out their NCAA Basketball March Madness brackets. At the start of 2017 March Madness, ESPN recorded 18.8 million basketball brackets. The first step in having a perfect bracket is picking the first round correctly.
Unfortunately, a lot of people cannot predict the future; well, no one can. Last year, only 164 brackets were perfect from the first round, which is less than 0.001% of the overall brackets. A lot of brackets get busted when lower-seeded teams upset the top-seeded teams.
Since the NCAA expanded in 1985 to 64 teams, at least eight upsets are recorded every year. If you want a winning bracket, you need to pick at least one lower-bracket team to get an upset win against the top teams. In this article, we will discuss computer model predictions of March Madness. Either you have a lot of resources available or not, predicting a winning bracket is possible.
Humans are error prone
It is not easy to know which of the lower-bracket teams will score an upset. Let us say you will decide between the number ten seeded and the number seven seeded teams. The lower ranked team has scored an upset in the tournament this year and even appeared in the final four in the past.
The higher ranked team is a team that is under the radar, meaning it does not receive that much national coverage. Most probably, casual fans never heard of that team. When given a choice, which side will you choose? If you picked the lower-bracket teams in 2017, you would have been wrong since the 10th-seed Virginia Commonwealth University won over Saint Mary’s of California.
Thanks to the recent bias, humans are prone to errors and can be tricked into using the latest observation in decision making. Recent bias is a kind of bias that is used in the picking process. Some people are biased in choosing their home team, whether they are playing good or bad.
There are also people who identify with individual players and want them to succeed. All of these biases can influence your decision-making process when negatively choosing your bracket. Even experts fall into recent biases.
Modeling upsets
Machine learning can help defend against these kinds of traps. Using machine learning, mathematicians, statisticians, as well as computer scientists train the machines to make an accurate prediction as possible by teaching it to learn from previous data and information.
This kind of approach has been used in many fields including medicine, marketing, and of course, sports. Machine learning methods can be compared to the black box of an airplane. The first thing you need to do is to feed the information, setting the dials on the plane’s black box.
Once they are calibrated, the algorithm will read the data that are fed, compare it with the previous data, analyze it, and gives accurate-as-possible predictions. Machine learning uses the data from 2001 to 2018 first-round teams to set the black boxes’ dials.
When the data are fed, it gives a 75% success rate prediction. It gives the experts a lot of confidence to analyze previous data, instead of trusting the predictions to luck. Because of this, it gives people a better chance to predict a winning bracket.
