Setting up your first Pine Sports Predict model can seem intimidating but PineHelp is here to help you find an edge over sportsbooks with AI powered predictions. With more than 50 different statistics to use as criteria, how do you know which ones to choose? In this article, I'll get you started in constructing a useful NFL prediction model for moneyline bets that will help you build a profit this year!
A good first step for Predict actually involves Explore+, and it's called 'correlate'. If you don't have access to Explore+, make sure to go into your account and use code: DISCORD to gain access to all features. Start by going to the Pine-Sports home page, click on 'Explore', then 'NFL' and lastly 'Game Stats'. On this screen, I would click the Season Year column header and set to equal the 2021-2022 season. After you've done that, or skip that if you want the last 5 seasons of data to be included, click on 'Correlate' in the tool bar.
.png)
Once you reach the Correlation page, you'll see a lot of random green and red boxes underneath two dropdowns. Click the first dropdown and select 'Final Score Spread'. While we aren't building a spread model, this is the option we want to use to help determine desired criteria for moneyline outcomes.
As you scroll through Correlate mark down which factors you see that are strongly correlated with final score spread. A strong correlation can be positive or negative, but the strongest correlations are the one's furthest from zero. If we determine variables with strong correlations to final score spread, and use those when constructing our predict model, the tool will use the value of those variables to help predict outcomes. Now it's time to head to 'Predict' to start building your model!
.png)
Once you get to the NFL Predict page, click 'Build a New Model' to get started. Put in the name for your model, set to predict the 'Winner', and now the fun starts. The next option will be your first decision to make, how many seasons of past data do you want to train your model off of? There is not necessarily a correct answer here, you can use up to 4 seasons or as little as 1 season. I generally stick to 2-3 seasons for all sports because there are so many changes across leagues over a 4-year period. Next, you'll have to decide a rolling average for the model to compare teams with. For this option, you can use as many as 16 games which would be the entirety of last season. I like to stick to around 8-10 personally, but again there is no right answer and you should play around with these settings to find what works best for your specific model. The reason I use 8-10 is because a two month period in NFL should not be swayed too much by one outlying performance.
.png)
Once you have the seasons of training data and the number of games for rolling average set, put in the criteria that you decided on from the 'Correlate' page and click 'Build Model'. After a minute or two, your model will be completely built and you can go see how it looks! Back at the Predict page, click 'Winner' this time to view your Winner model score. If you click the green icon next to your model score, you can do further analysis to see where it is most effective along with the overall confidence range that you can expect throughout the season. In my case, a confidence level of 65% or higher was correct 70% of the time out of 224 predictions. In order to see who your model predicts for this week, just click on 'Get Latest Predictions'!
.png)
You have successfully built a Pine Sports Prediction model and can replicate this process for Spread and Total Score predictions in NFL, as well as NBA, NHL, and MLB predictive models. For NBA, NHL, and MLB you will want to play around more with season and rolling average. What work for one sport may not necessarily be as effective for another.
Setting up your first Pine Sports Predict model can seem intimidating but PineHelp is here to help you find an edge over sportsbooks with AI powered predictions. With more than 50 different statistics to use as criteria, how do you know which ones to choose? In this article, I'll get you started in constructing a useful NFL prediction model for moneyline bets that will help you build a profit this year!
A good first step for Predict actually involves Explore+, and it's called 'correlate'. If you don't have access to Explore+, make sure to go into your account and use code: DISCORD to gain access to all features. Start by going to the Pine-Sports home page, click on 'Explore', then 'NFL' and lastly 'Game Stats'. On this screen, I would click the Season Year column header and set to equal the 2021-2022 season. After you've done that, or skip that if you want the last 5 seasons of data to be included, click on 'Correlate' in the tool bar.
Once you reach the Correlation page, you'll see a lot of random green and red boxes underneath two dropdowns. Click the first dropdown and select 'Final Score Spread'. While we aren't building a spread model, this is the option we want to use to help determine desired criteria for moneyline outcomes.
As you scroll through Correlate mark down which factors you see that are strongly correlated with final score spread. A strong correlation can be positive or negative, but the strongest correlations are the one's furthest from zero. If we determine variables with strong correlations to final score spread, and use those when constructing our predict model, the tool will use the value of those variables to help predict outcomes. Now it's time to head to 'Predict' to start building your model!
Once you get to the NFL Predict page, click 'Build a New Model' to get started. Put in the name for your model, set to predict the 'Winner', and now the fun starts. The next option will be your first decision to make, how many seasons of past data do you want to train your model off of? There is not necessarily a correct answer here, you can use up to 4 seasons or as little as 1 season. I generally stick to 2-3 seasons for all sports because there are so many changes across leagues over a 4-year period. Next, you'll have to decide a rolling average for the model to compare teams with. For this option, you can use as many as 16 games which would be the entirety of last season. I like to stick to around 8-10 personally, but again there is no right answer and you should play around with these settings to find what works best for your specific model. The reason I use 8-10 is because a two month period in NFL should not be swayed too much by one outlying performance.
Once you have the seasons of training data and the number of games for rolling average set, put in the criteria that you decided on from the 'Correlate' page and click 'Build Model'. After a minute or two, your model will be completely built and you can go see how it looks! Back at the Predict page, click 'Winner' this time to view your Winner model score. If you click the green icon next to your model score, you can do further analysis to see where it is most effective along with the overall confidence range that you can expect throughout the season. In my case, a confidence level of 65% or higher was correct 70% of the time out of 224 predictions. In order to see who your model predicts for this week, just click on 'Get Latest Predictions'!
You have successfully built a Pine Sports Prediction model and can replicate this process for Spread and Total Score predictions in NFL, as well as NBA, NHL, and MLB predictive models. For NBA, NHL, and MLB you will want to play around more with season and rolling average. What work for one sport may not necessarily be as effective for another.
Let us know by commenting below 👇
Are we able to see how confident the model is for that day's predictions?