Yes, ladies and gentlemen, we are on a two-game winning streak. This has been a successful run of games that can be attributed to a few factors that will be discussed later on in the article. In any fantasy sport, all decisions are data-driven, however, fine-tuning the analysis, interpreting it and extrapolating it into the future is often not a trivial task. As we are at the halfway point of the season, in this article we will take a look at what went right, what we could have done better and what has led us to win 3 out of our 4 four matchups.
To start off let’s look at what we could have done better. The scientific method often required assumptions to be made. In this case, we made two key assumptions: firstly, last season’s trends would carry into this season. We made this assumption because there was no data on the current season. The second assumption we made was that analysing only the quarterbacks, running backs, tight ends and wide receivers would be sufficient to project a winning team.
This means we left out defensive lines and kickers in our analysis. This was done to reduce the complexity of our analysis and data collection process. Looking at the first assumption, this was a necessary assumption due to what’s referred to as the cold start problem. In order to make good predictive analytics you need data, but early in the season there just isn’t enough data to make good predictions. This lack of data and hence no ability to find trends is what’s referred to as the cold start problem. There are many ways around this, one way is simply to guess and optimise as more data comes in. This is a valid solution but not ideal in this context. The route we chose was to use previous seasons data until enough of this season’s data was gathered to find meaningful trends.
This method has high variability and is inherently error-prone in the beginning, but over time converges to a workable model. This is partly the reason we did not perform very well in the beginning weeks.The second assumption, omitting kickers and defensive teams, was certainly erroneous. When recapping matchups we’d lost, it was very evident that these positions had a significant contribution to our opponent’s totals, sometimes up to 10% of the total score. Due to the nature of NFL fantasy, it’s difficult to get trades.
This means every successful trade has to be a high performer and therefore any errors in assumptions are more impactful.Now that we’ve taken a look at what went wrong, let’s look at what we did well. In the last two weeks, we started including kickers and defensive teams in our analysis. Fortunately, our defensive team had a favorable matchup against the Chicago Bears in week 7. We decided to keep them and this paid off massively as they attained a score of 18, which was 14% of our total. We also analysed the following position for the weeks to follow (QB). We are looking into possibly picking up Tua Tagovailoa (Miami Dolphins, QB) or Matt Ryan (Atlanta Falcons, QB) as Aaron Rodgers is on Covid-19 Protocol and our leading QB (Matthew Stafford) is listed as questionable. Because these players are available in free agency, it means that they are not typically rostered on a high percentage of teams.
Therefore, we will be sacrificing a few points if Stafford is not listed as “healthy” come Sunday night — we will hold our breath until then! (Fingers crossed!).We now have a lot of data on player performances and were able to make better predictions. This has allowed us to win back-to-back games, picking up three out of our last four. Data science is an iterative process by nature, so we will continue to optimise and improve the analysis and hopefully continue our current form.
However, if there’s one thing guaranteed in Fantasy Sports, it’s that nothing is guaranteed, though by using the power of data and analysis, we’re certainly in a good position to “soldier on” to fantasy glory.A lot of teams have gigabytes of GPS and video meta-data with no easy way to interpret it.
Through AthStat, teams can easily generate various visualizations giving a more in-depth breakdown into team performance. Analyzing your data is as easy as dragging and dropping it into the platform. The platform automatically aggregates the previous data you’ve uploaded keeping all your visualizations up to date.
Get all the latest news on new features that are upcoming and the most awesome blog articles.
Join our newsletter for updates. Read our Terms