After conducting numerous tests based on the NBA's classical advanced statistics, we have decided to create our own performance metric: the Impact Performance. This metric can be viewed as a modernized and more representative PER (Player Efficiency Rating). For each type of prediction, we use a different machine learning model, but the Impact Performance is always the first feature that the algorithm is built upon.
For each game, we take into account the Impact Performance of the players who will be on the court that day, as well as the team's overall Impact Performance and more traditional statistics (TS%, rebound presence, TOV, etc.). We then apply a neural model that considers the player's role in the team, and from this, we deduce a probability of victory.
Our goal was to outperform the bookmakers, and we have succeeded. On average, we have achieved a success rate of over 63% over the past two seasons.
However, it is important to note some limitations to consider if you wish to bet on our predictions.
The model is only applicable after about fifteen games per team and only shows its real effectiveness after about thirty games.
The trade deadline periods and the last days of the regular season are also times when the model tends to underperform.
Lastly, our model poorly integrates the back-to-back factor, so we advise you to bet cautiously on these matches.
To predict the winner of the MVP, we also use the Impact Performance, which we weight by the player's team win ratio. This method has allowed us to predict the MVP for the last twelve seasons with a success rate of 92%. The only miss was the 2016-2017 season, during which the algorithm predicted James Harden as the MVP, but it was Russell Westbrook who ultimately won the title by averaging a triple-double (a metric that our algorithm does not consider).
| Season | Prediction | Winner |
|---|---|---|
| 2011-2012 | LeBron James | LeBron James |
| 2012-2013 | LeBron James | LeBron James |
| 2013-2014 | Kevin Durant | Kevin Durant |
| 2014-2015 | Stephen Curry | Stephen Curry |
| 2015-2016 | Stephen Curry | Stephen Curry |
| 2016-2017 | James Harden | Russell Westbrook |
| 2017-2018 | James Harden | James Harden |
| 2018-2019 | Giannis Antetokounmpo | Giannis Antetokounmpo |
| 2019-2020 | Giannis Antetokounmpo | Giannis Antetokounmpo |
| 2020-2021 | Nikola Jokic | Nikola Jokic |
| 2021-2022 | Nikola Jokic | Nikola Jokic |
| 2022-2023 | Joel Embiid | Joel Embiid |
Predicting the Rookie of the Year is even simpler because we do not apply any weighting here. We take the Impact Performance of rookie players and compare them. Over the last twelve seasons, we have achieved a success rate of 75%.
| Season | Prediction | Winner |
|---|---|---|
| 2011-2012 | Kyrie Irving | Kyrie Irving |
| 2012-2013 | Damian Lillard | Damian Lillard |
| 2013-2014 | Michael Carter-Williams | Michael Carter-Williams |
| 2014-2015 | Nerlens Noel | Andrew Wiggins |
| 2015-2016 | Karl-Anthony Towns | Karl-Anthony Towns |
| 2016-2017 | Dario Saric | Malcolm Brogdon |
| 2017-2018 | Ben Simmons | Ben Simmons |
| 2018-2019 | Luka Doncic | Luka Doncic |
| 2019-2020 | Ja Morant | Ja Morant |
| 2020-2021 | LaMelo Ball | LaMelo Ball |
| 2021-2022 | Evan Mobley | Scottie Barnes |
| 2022-2023 | Paolo Banchero | Paolo Banchero |
The prediction of the Defensive Player of the Year is the least reliable and is based on more traditional NBA statistics.
Since 2012, we have only achieved a success rate of 58%, but upon reviewing the list, you will notice that the predictions from our models indicate excellent defenders.
| Season | Prediction | Winner |
|---|---|---|
| 2011-2012 | Kevin Garnett | Tyson Chandler |
| 2012-2013 | Tim Duncan | Marc Gasol |
| 2013-2014 | Joakim Noah | Joakim Noah |
| 2014-2015 | Draymond Green | Kawhi Leonard |
| 2015-2016 | Kawhi Leonard | Kawhi Leonard |
| 2016-2017 | Draymond Green | Draymond Green |
| 2017-2018 | Rudy Gobert | Rudy Gobert |
| 2018-2019 | Giannis Antetokounmpo | Rudy Gobert |
| 2019-2020 | Giannis Antetokounmpo | Giannis Antetokounmpo |
| 2020-2021 | Rudy Gobert | Rudy Gobert |
| 2021-2022 | Robert Williams | Marcus Smart |
| 2022-2023 | Jaren Jackson Jr. | Jaren Jackson Jr. |
The prediction of the Sixth Man of the Year also relies on Impact Performance, but it is applied only to bench players. We have achieved a success rate of 58% over the last twelve seasons, but we have observed a much better prediction rate over the last six seasons (83%).
| Season | Prediction | Winner |
|---|---|---|
| 2011-2012 | James Harden | James Harden |
| 2012-2013 | J.R. Smith | J.R. Smith |
| 2013-2014 | Manu Ginobili | Jamal Crawford |
| 2014-2015 | Josh Smith | Lou Williams |
| 2015-2016 | Hassan Whiteside | Jamal Crawford |
| 2016-2017 | Lou Williams | Eric Gordon |
| 2017-2018 | Lou Williams | Lou Williams |
| 2018-2019 | Montrezl Harrell | Lou Williams |
| 2019-2020 | Montrezl Harrell | Montrezl Harrell |
| 2020-2021 | Jordan Clarkson | Jordan Clarkson |
| 2021-2022 | Tyler Herro | Tyler Herro |
| 2022-2023 | Malcom Brogdon | Malcom Brogdon |
Last but not least, the prediction of the playoffs winner. We use the team's Impact Performance, but also its ability to limit the opponent's IP, the TS% (True Shooting Percentage), winshare, etc.
Conducting a benchmark is more complicated because we do not announce a single winner but instead provide a victory probability for each team.
However, let's note a few points:
NBAprediction