Casey Cooper

A showcase of my personal projects as well as various data science techniques. My resume can be found on the About Me page.

Sports Betting ARM Conclusions

‘Under’

Strong Wind appears in five of the top 10 rules when sorting by confidence. While the support is very low for these rules, the confidence is equal to 1. Strong Wind is not super common at all, so it is expected for the support to be low. These rules also have lift close to 2, so the association is highly correlated. This means that Strong Wind (winds between 25-31mph) could be associated with the under occurring. This is worth looking further into. The following rule will show if strong winds are associated with the under occurring.

  • {Strong Wind} => {Under} 0.004220779, 0.8125, 0.005194805, 1.577372, 26

The lift is approximately 1.58, which means there is a strong association between strong wind and the under occurring. The confidence is also fairly high at approximately 0.81. While, strong wind occurring is pretty rare, as shown by support, when it does occur it is pretty strongly correlated with the under occurring.

Mike Carey (a referee) is in 7 of the top 10 rules when sorting by support. These rules still have confidence over 0.9. These rules also have a lift over 1.7, so the association is highly correlated. The total home offense rank between 1 and 5 is also in 5 of the top 10 rules when sorting by support. This means that Mike Carey and total home offense rank between 1 and 5 are also looking further into, as they could be associated with the under occurring.

  • {Mike Carey} => {Under} 0.02045455 0.5860465 0.0349026 1.137739 126

The lift is approximately 1.14, which means there is an association between Mike Carey and the under occurring. The confidence is only 0.59. However, bets only have to be correct 52.4% of the time to be profitable, so this confidence is acceptable. So, based on the above rule it appears that Mike Carey is moderately correlated with the under occurring. Unfortunately, upon further research, Mike Carey has retired from his referee position in the NFL.

  • {Total Home Offense Rank of 1-5} => {Under} 0.08814935 0.4909584 0.1795455 0.953137 543

The lift is approximately 0.95, which means there is a negative association between the home team offense being ranked in the bottom five of the league and the under occurring. This means that the home team offense being ranked in the bottom five alone is not associated with the under. Instead it needs to be combined with other categories to correlate with the under.

All of these top 10 rules are worth noting and keeping, since the confidence and lift are high. This means that if these specific rules ever match up to future NFL game the Under will probably happen. Since the support is low, it means that these specific rules won’t match up often though.

‘Over’

Non-divisional games show up in 7 of the top 10 rules when sorting by confidence or support. Since, the confidence is higher than 0.9 and lift is higher than 1.94 there is a strong association and confidence between these rules containing a non-divisional game and the over occurring. This is definitely something to look further into.

  • {non_div_game} => {Over} 0.3097403 0.4954557 0.6251623 1.021763 1908

The lift is approximately 1.02, which means there is a slight association between a non-divisional game and the over occurring. The confidence is only approximately 0.50, which is too low to be useful to sports bettors. So, unfortunately it appears that just having a non-divisional game alone isn’t enough to be correlated with the over. Most likely it slightly helps contribute to the over occurring when combined with other categories.

All of these top 10 rules are also worth noting and keeping, since the confidence and lift are high. This means that if these specific rules ever match up to future NFL game the Over will probably happen. Since the support is low, it means that these specific rules won’t match up often though.

Overall

Overall, Association Rule Mining found many rules that have a confidence over 0.9. However, these rules don’t occur very often, so they usually won’t be applicable for sports bettors on a weekly basis during the NFL season. Association Rule Mining also found many rules that do appear more often, with a support of at least 0.1 and a confidence of at least 0.55. A support of 0.1 is still somewhat low to be applicable on a weekly basis though. So, Association Rule Mining was successful in finding certain rules that correlate with a winning bet on the over or under at least 52.4%. However, the rules are somewhat rare, which means sports bettors won’t be able to use these rules for any game on a weekly basis. All of the useful rules can be found on the ARM results page.