Konosa Pro Football Picks
This is the website for the alpha version of the pro football game picking system from Konosa. Konosa is a website dedicated to Big Data projects, and this particular section of it is dedicated to the specific goal of attempting to predict NFL games.
First word of advice: if you're planning to gamble on pro football games, please have sense enough to not treat the data that we provide here as a unitary gospel of how you should bet. This is the very first published version of these results, and we're still in the process of validating our picks against live games. We're a long way from being able to state categorically that we can pick NFL games. That said, we're all capable of making our own decisions. No whining about the picks is allowed.
This is still an alpha version of the system. While we're comfortable enough with the results that we've seen to publish, our primary goal at this stage is to validate the system and tweak and fine tune it. For example, we're still a way off from properly quantifying the impact that unreliable quarterbacks have on games. It's an idea that's fairly easy to appreciate in the abstract, but it's much harder to nail down. If you have any doubts about that issue, consider the issue of figuring out how to pick games where Eli Manning is the QB. Yeah . . . you get the idea pretty quickly.
Within the system itself, our goal is to pick teams against the moneyline. The moneyline represents the simplest version of a sports bet to treat as a straightforward financial instrument. We don't do point spreads, the over-under or any other variant of sports game picking.
The moneyline that we're using is the best available line from 5dimes within two to 24 hours of the start of the game. If there is a reduced juice line available, that is the line that will be used. That said, moneylines change. DO NOT RELY UPON OUR MONEYLINES. Ever. Period.
Current State of the Site
At this stage, we're able to publish limited information about the games. All previous games that we publish will include links to the NFL Game Center data for the individual game in order to permit visitors an opportunity to get a sense of the statistics that fed the model.
The website is not being currently monetized by any means except advertising. While we would prefer in an ideal world to keep it this way, users should be prepared for the day that we decide to go to a paid, limited, or gated model. There will be no complaining about the day that occurs.
For now, you have a free peek at the predictions that we're trying to make. We would appreciate all of the feedback that's possible, and we'll try to keep in mind who the folks were who helped us improve the website if we ever go to a membership model and offer them some type of discount or reward for their assistance during the alpha testing and validation phases.
About the Models Themselves
We currently utilize two models. These models are called "Model 1" and "Model 3". To answer the obvious question, there is a Model 2, but it was an abject failure and could not be used to predict games in any way.
Model 1 is a fixed model. It assumes that the teams are evolving, but that the game of pro football itself is not. It does a fairly good job of picking games, but it is mostly included in order to throw up red flags about the games the Model 3 picks.
Model 1 is also beneficial because it permits us to provide a clearly linguistic narrative about each game. The groupings of games are fixed, and therefore we have a solid idea of what those games are like. We can easily state that the major of games that are labeled, for example, "Cluster 3" away games are games where the score shifted back and forth and both teams made mistakes while also moving the ball. This can be very helpful when simply trying to understand what exactly happened during a game.
Model 3 is an evolutionary model. It assumes that not only are the teams themselves changing, but that the game of professional football played within the NFL itself is also changing. This means that the clusters of games should be expected to change every week, due to both the randomization applied to the data and the actual changes that are occurring within the game of football.
We also employ an "accelerated" version of Model 3. The acelerated Model 3 attempts to predict where a team's performance might be headed. This can be helpful since a large split between the general Model 3 and the accelerated Model 3 can be indicative that a team is experiencing a significant improvement or decline in its game. A team that's experiencing a significant decline, such as the Houston Texans between the 2012 and 2013 seasons, will see a significant gap appear that reflects the acceleration of the team's decline, where a 55% probability in the general model might appear as a 51% probability in the accelerated model. Likewise, an improving team, such as the 2014 Dallas Cowboys will see a gap form headed upward. For example, the prediction gap for the Cowboys' week 7 home game against the Giants showed a 51% probability of a win in the general versus a near 60% probability. This gap indicates that if those two teams were to meet down the road, such as in a playoff game, we'd expect the Cowboys to have an even better chance of winning the next time. Also, it implies that the team's current chances are probably better than the general model is stating.
We will make our best effort to keep the model as simple and readable as possible. If you feel that something is either unclear or too complicated, please let us know.
We hope you enjoy following our predictions, and we look forward to discussing them with you.
Sites from KonosaMain Site
NCAA Football Picks
Follow Us!On Twitter:
The Konosa blog