Hope you all can get something useful from this and, as always, best of luck everyone!
Updated projections with latest model on 5 Feb 2021. Latest model includes Corsi, Fenwick, & High Danger stats.
Team Consistency Update
Here is a look how consistent the Hockey Teams have been so far.
What this chart is showing is each teams Variance, Standard Deviation (Std Dev), and Average score. The list is ordered from lowest team variance to most team variance. The variance is how wide spread the data is, a team that scores between 1 and 2 goals every night will have very low variance, whereas a team that scores anywhere from 0 to 6 on a given night will have very high variance. The Std Dev is the square root of the variance and is a good measure for how consistent a team is… NOT how good or bad a team is, but how consistent they are. Std Dev shows the amount a teams score typically deviates from the average on a given night. The Red Wings score (typically) will deviate about 0.69 goals from there average on a given night, where the Canucks score (typically) will deviate 3.78 goals from their average on a given night. Obviously the lower the Std Dev the easier it is for my models to project the score and provide higher probabilities.
The overall records updated, but the output for yesterdays games didn’t save. I am extremely busy and just want to get projections out, I don’t have time to go back and re-grade everything manually just for the images. ATS did well (9-3 for all), ML was ~.500 (depending on the model), O/U was just under ~.500 for most models.
Top Performing Models
Based on Average Win % between ATS, Moneyline, and O/U.
Updated after todays games.
- kNN – 60%
- Random Forest – 60%
- SVM – 58%
- Neural Network – 57%
- Ada Boost – 56%
- Linear Regression – 53%
Summary of Projections
1. k-Nearest Neighbor (kNN)
The lazy learner (That’s what my grad school prof called it). Nearest Neighbor has performed best with 52 Neighbors.
How to read the projections: The model type is at the very top. Matchups are denoted by the alternating white/green pairings, away teams are on top, home teams on bottom. The model analyzes the matchup and projects the home and away teams score (Proj Score). The difference between the home teams score and the away teams score give us what the model says the line should be (Proj Line). The “Line” column is what the Vegas line is at the time I run the model. The “Line Diff” is the difference in the projected line and the Vegas line. A positive line diff means the projected line is in the away teams favor compared to the Vegas line, negative means the projected line is in the home teams favor. The “Cover Prob” uses a normal distribution and the teams variance to project each teams probability to cover the listed Vegas line. Same thing for totals, “Proj Total” is the sum of the projected scores, “Line Total” is the listed Vegas total for the game, the difference between the two, and the probability to go over or stay under denoted by “O” and “U”. The “ML prob” is the probability of a team to cover the moneyline, or the outright winner.
If you don’t understand what you are looking at, I recommend reading my post about betting tips. The percentages show the betting Edge, which is the cover prob (from above) minus the implied probability (-110 odds implied prob is 52.4%). If a team has a 92.4% chance to win and a 52.4% implied probability (or -110 odds), the Edge is 40%. The Edge alone doesn’t mean you should blindly bet it. To quantify, >35% is great value, 20-35% is really good, 10-20% is decent, <10 is ok value, blanks are negative value.