human computer interactive learning - prediction
http://clockworkchaos.com/project7/?q=taxonomy/term/34
enNFL Kickers Now 10x More Likely to Miss the Extra Point!
http://clockworkchaos.com/project7/?q=nfl_pat
<!-- google_ad_section_start --><div class="field field-name-body field-type-text-with-summary field-label-hidden"><div class="field-items"><div class="field-item even" property="content:encoded"><p>Recent NFL rule change has moved the point after a touch (PAT) down from the 2 yard line back to the 15 yard line. So far the chances of making the extra point is down to 95%. That may seem high, but it is actually 4.5% lower!<br /></p><center>
<iframe height="500" width="720" scrolling="no" frameborder="0" src="http://clockworkchaos.com/projects/nfl/pat/byTypeWithPrediction.html">
<p>Your browser does not support frames, consider using Mozilla/Chrome.</p>
</iframe><p></p></center><br />
Over the last five years the PAT completion rate has remained mostly unchanged. The timing of the rule change coincides with a decrease chance of completing the PAT. This is not due to a reduction in the skill of our NFL kickers. Actually, we can see that NFL kickers have been consistently improving in their ability to complete field goals. Overall, NFL kickers are 3.3% more likely to make a field goal in 2015 than they have been in previous five seasons. Kickers are consistently increasing their ability to complete field goals each year.<br /><img align="middle" width="500" src="http://clockworkchaos.com/projects/nfl/pat/fgTrend.PNG" alt="This chart shows the probability of an NFL completing a field goal from 2010 through 2015. (Scaled 0-100)" /><br />
The chart above shows the probability of an NFL kicker completing a field goal. The chart below shows the same chart zoomed to the scale of the data:<br /><img align="middle" width="500" src="http://clockworkchaos.com/projects/nfl/pat/zoom_fgTrend.PNG" alt="This chart shows the probability of an NFL completing a field goal from 2010 through 2015. (Scaled 75-100)" /><br />
We could argue that their increase in performance is due to improved coaching. The kickers are not asked to attempt kicks which the coach is not confident that they can complete. To test this look at completions by yardage by year:<br /><img align="middle" width="500" src="http://clockworkchaos.com/projects/nfl/pat/successTable.PNG" alt="What is the point of all of these normal distributions?" /><br />
The table above shows the probability that an NFL kicker will complete a field goal by year and by 20-yard bracket. It looks like these kickers are still doing as well across the various yard-based buckets. So if we compare the field goal completion rate to the PAT completion rate, what do we notice?<br /><img align="middle" width="500" src="http://clockworkchaos.com/projects/nfl/pat/byType_simple.PNG" alt="This chart shows the probability of an NFL completing a field goal alongside the probability of completing a PAT from 2010 through 2015. (Scaled 0-100)" /><br />
Here is the same chart zoomed to the scale of the data. I also added a highlighted point that indicates what the expected completion rate should be for PATs.<br /><img align="middle" width="500" src="http://clockworkchaos.com/projects/nfl/pat/byTypeWithPrediction_simple.PNG" alt="This chart shows the probability of an NFL completing a field goal alongside the probability of completing a PAT from 2010 through 2015. It also shows the expected 99.5% PAT completion rate. (Scaled 75-100)" /><br />
The extra point was gone from being a given (99.56%) to having a 19/20 chance. Perhaps this is why the NFL felt motivated to change the rule. Is it a large enough change? If you do the math, it turns out that kickers are now 10x more likely to miss the extra point! Wow- time to adjust our strategies. Will more teams start trying to convert on a touch down?
</div></div></div><!-- google_ad_section_end --><span class="vocabulary field field-name-field-tags field-type-taxonomy-term-reference field-label-above"><h2 class="field-label">Tags: </h2><ul class="vocabulary-list"><li class="vocabulary-links field-item even" rel="dc:subject"><a href="/project7/?q=taxonomy/term/31" typeof="skos:Concept" property="rdfs:label skos:prefLabel">NFL</a></li><li class="vocabulary-links field-item odd" rel="dc:subject"><a href="/project7/?q=taxonomy/term/80" typeof="skos:Concept" property="rdfs:label skos:prefLabel">Analytics</a></li><li class="vocabulary-links field-item even" rel="dc:subject"><a href="/project7/?q=taxonomy/term/47" typeof="skos:Concept" property="rdfs:label skos:prefLabel">analysis</a></li><li class="vocabulary-links field-item odd" rel="dc:subject"><a href="/project7/?q=taxonomy/term/79" typeof="skos:Concept" property="rdfs:label skos:prefLabel">Predictive Analytics</a></li><li class="vocabulary-links field-item even" rel="dc:subject"><a href="/project7/?q=taxonomy/term/34" typeof="skos:Concept" property="rdfs:label skos:prefLabel">prediction</a></li></ul></span>Wed, 30 Sep 2015 05:04:28 +0000rakirk47 at http://clockworkchaos.com/project7http://clockworkchaos.com/project7/?q=nfl_pat#commentsHow to Pick the Perfect Super Bowl Squares
http://clockworkchaos.com/project7/?q=superbowl-squares
<!-- google_ad_section_start --><div class="field field-name-body field-type-text-with-summary field-label-hidden"><div class="field-items"><div class="field-item even" property="content:encoded"><p>Super Bowl Squares are a popular past time where sports fans guess (bet on) the outcome of one of game. But what numbers should people chose?<br /><img align="center" width="400" src="http://clockworkchaos.com/projects/benford/historical_odds.PNG" alt="Historical numerical frequencies for NFL games." /><br />
The way this bet works, a user must chose what the last digit of the final score will be for both teams. For example, when the Seahawks beat the Broncos with a score of 17-24 (this is my actual prediction), then the person with the square at the intersection of 7 to 4 would win. I'm optimistic since I know the Vegas spread favors the Broncos by roughly 3 points. So if the final score were 24-17 Broncos, then the person with the [4,7] square would win. The graph above shows the most likely outcomes from all 14,000+ historic games. Going up and down are the odds for the scores from the winning team. Going left to right visualizes the scores from the losing team. Dark blue indicates an outcome three times as likely as the average outcome. Teal indicates average likelihood. Yellow Indicates below normal odds. These colors are not colorblind safe, but they do go well with Seakhawks colors.<br /><br /><br />
Notice several things: there are patterns in the numbers<br /></p><center><img align="center" width="400" src="http://clockworkchaos.com/projects/benford/historical_ranks.PNG" alt="Historical numerical ranks for NFL games." /></center><br />
The purpose of the chart above is to help make the first chart clearer. In order to make decision making more meaningful, this chart forces some distinct between similar values through ranking each from most likely to least likely. This should make it even clearer to the reader which positions are more or less likely to win. Here dark blue represents the number 1 position and bright white represents the worst position. Again, similar patterns are present as in the first chart but there is now increased variation.<br /><br /><br />
So what? Perhaps others have already analyzed the Super Bowl squares problem. What frustrates me is that none of these predictions seem to account for a common and important principle. Mathematicians have known about this secret for over a hundred years. Simply put, the smaller the number is, the more likely it is to occur. So, when it comes to the perfect squares problem this is important. The number 0 will occur more often by chance. Then 1 will be the next most common and so on. Actually the charts below summarize this effect.<br /><center><img align="center" width="225" src="http://clockworkchaos.com/projects/benford/benford_table.PNG" alt="Natural distribution of the trailing digit of a two-digit number in a table." /><img align="center" width="400" src="http://clockworkchaos.com/projects/benford/benford_odds.PNG" alt="Natural distribution of the trailing digit of a two-digit number in histogram chart." /></center><br />
Armed with this new knowledge, recreate the previous chart to account for these likelihoods.<br /><center><img align="center" width="400" src="http://clockworkchaos.com/projects/benford/benford_ranks.PNG" alt="Natural distribution of the trailing digit of a two-digit number." /></center><br />
Notice that many of the same trends still occur. However, the region in the bottom-right is now more important than the top-right. Specifically the numbers [7,6],[7,7], and [8,7] are now more common than [0,0],[1,0], and [1,1]. This represents a quantifiable advantage useful for decision-making. Actually, the image below shows the top 10 squares to consider choosing both before and after accounting for the Benford frequencies.<br /><center><img align="center" src="http://clockworkchaos.com/projects/benford/squares_top10.PNG" alt="Top 10 picks before and after Benford normalization." /></center><br />
Notice the how the number one spot changes from [0,7] to [4,7]. So, if the most likely outcome is supposed to be [4,7], what will the full score be? Will it be 14-17, 17-24, or 7-14, etc? And which team will win? It turns out that mono-frequency football analysis favors the winning team having a score starting with a 2 and the losing team having a score staring with a 1. This suggests a score such as 24-17. (I can post this if there is interest). In terms of which team will win... Lynch and Sherman are two of the reasons why I choose to make this prediction in favor of the Seahawks.<br /><br /><br />
Data Sources: <a href="http://www.pro-football-reference.com/boxscores/game_scores.cgi/">Football Scores</a>, <a href="http://www.jstor.org/stable/2369148?seq=2">Frequency of Natural Numbers (Benford's Law).</a>
<div class="facebook-like">
<iframe src="http://www.facebook.com/plugins/like.php?href=http://clockworkchaos.com/project7/node/26&layout=standard&show_faces=false&width=400&action=like&font=arial&colorscheme=light&height=35" scrolling="no" frameborder="0" style="border:none; overflow:hidden; width:400px; height:35px;" allowtransparency="true"></iframe></div>
<div class="fb-comments">
<iframe src="http://www.facebook.com/plugins/comments.php?href=http://clockworkchaos.com/project7/node/26&permalink=0" scrolling="yes" frameborder="0" style="border:none; width:720px; height:400px; scrolling=" true="" allowtransparency="true"></iframe></div>
</div></div></div><!-- google_ad_section_end --><span class="vocabulary field field-name-field-tags field-type-taxonomy-term-reference field-label-above"><h2 class="field-label">Tags: </h2><ul class="vocabulary-list"><li class="vocabulary-links field-item even" rel="dc:subject"><a href="/project7/?q=taxonomy/term/34" typeof="skos:Concept" property="rdfs:label skos:prefLabel">prediction</a></li><li class="vocabulary-links field-item odd" rel="dc:subject"><a href="/project7/?q=taxonomy/term/43" typeof="skos:Concept" property="rdfs:label skos:prefLabel">odds</a></li><li class="vocabulary-links field-item even" rel="dc:subject"><a href="/project7/?q=taxonomy/term/40" typeof="skos:Concept" property="rdfs:label skos:prefLabel">Indexing</a></li><li class="vocabulary-links field-item odd" rel="dc:subject"><a href="/project7/?q=taxonomy/term/31" typeof="skos:Concept" property="rdfs:label skos:prefLabel">NFL</a></li></ul></span>Thu, 30 Jan 2014 22:22:15 +0000rakirk26 at http://clockworkchaos.com/project7http://clockworkchaos.com/project7/?q=superbowl-squares#commentsWhen it comes to winning NFL games, what matters more than points per game?
http://clockworkchaos.com/project7/?q=nfl-yards
<!-- google_ad_section_start --><div class="field field-name-body field-type-text-with-summary field-label-hidden"><div class="field-items"><div class="field-item even" property="content:encoded"><p>The single most important variable is the total yards per game. This single factor is well balanced because it accounts factors such as the number of games played and the type of play.<br /></p><center>
<iframe height="360" width="720" scrolling="no" frameborder="0" src="http://clockworkchaos.com/projects/nfl/parallel/parallel.html">
<p>Your browser does not support frames, consider using Mozilla/Chrome.</p>
</iframe><p></p></center><br />
What factors are important for predicting total yards? The top three factors are: points per game, first downs per game, total scrimmage plays, yards per game, and first downs per game. So while points per game isn't the most important factor, it is still among the top predictive factors. It turns out that all of these top factors are also highly related to winning games. In fact, each of these explains more than 70% of the variance in points per game. This doesn't make sense, how can there be so many highly predictive features? Well, all of these top factors tend to rise and fall together. So, if all of these factors are intertwined. What's a good way to show the relationships between these factors?<br /><br /><br />
The chart above is a parallel coordinate plot of these top factors. Each line on this chart shows the outcome of a particular season for the Seahakws, the Broncos, and the Patriots. When you mouse over a line it will highlight the line revealing the total scores for all four variables. You can also re-arrange the order of the vertical lines by dragging and dropping them. Finally, try selecting all of the teams that had low total points per game and watch as it highlights related data. Notice how the low scoring points per game tend to also have fewer scrimmage plays, fewer first downs, and fewer yards per game.<br /><br /><br /><center><img align="center" width="450" height="450" src="http://clockworkchaos.com/projects/nfl/parallel/seahawk-stadium.JPG" alt="Century Link field in Seattle, WA." /></center><br />
What's the best way to show the relationship between these variables as it relates to teams and to players? The answer is to use the natural hierarchy between division, teams, and players to organize these information. Back to points per game for a second. This factor often used by sports fanatics to build fantasy football teams. Given the importance of points per game to choosing the right player for a team, what does it look like by player?<br /><center>
<iframe height="235" width="720" scrolling="no" frameborder="0" src="http://clockworkchaos.com/projects/nfl/hierarchy.html">
<p>Your browser does not support frames, consider using Mozilla/Chrome.</p>
</iframe><p></p></center><br />
The chart above shows total points per game across divisions, teams, and players. It is interactive. Blue bars indicate the presence of subcategories while grey bars indicate that a concept has no further subcategories. Click on a blue bar to zoom into a region and see a breakdown of data across related concepts. Click on the whitespace next tot he chart to to zoom out.<br /><br /><br />
Points per game may be an important factor for helping to pick the best offensive players for a given team. But don't forget about all of those defensive backs. They're big guys, it's not good to upset them. And obviously a good offense is only half the game. What method can account for the importance of defense to winning a game? What is relationship between these defensive factors and the top predicting factors? <i>I'll dig into these issues in more in the future.</i>
<div class="facebook-like">
<iframe src="http://www.facebook.com/plugins/like.php?href=http://clockworkchaos.com/project7/node/20&layout=standard&show_faces=false&width=400&action=like&font=arial&colorscheme=light&height=35" scrolling="no" frameborder="0" style="border:none; overflow:hidden; width:400px; height:35px;" allowtransparency="true"></iframe></div>
<div class="fb-comments">
<iframe src="http://www.facebook.com/plugins/comments.php?href=http://clockworkchaos.com/project7/node/20&permalink=0" scrolling="no" frameborder="0" style="border:none; overflow:hidden; width:720px; height:300px; scrolling=" true="" allowtransparency="true"></iframe></div>
</div></div></div><!-- google_ad_section_end --><span class="vocabulary field field-name-field-tags field-type-taxonomy-term-reference field-label-above"><h2 class="field-label">Tags: </h2><ul class="vocabulary-list"><li class="vocabulary-links field-item even" rel="dc:subject"><a href="/project7/?q=taxonomy/term/31" typeof="skos:Concept" property="rdfs:label skos:prefLabel">NFL</a></li><li class="vocabulary-links field-item odd" rel="dc:subject"><a href="/project7/?q=taxonomy/term/33" typeof="skos:Concept" property="rdfs:label skos:prefLabel">parallel coordinates</a></li><li class="vocabulary-links field-item even" rel="dc:subject"><a href="/project7/?q=taxonomy/term/30" typeof="skos:Concept" property="rdfs:label skos:prefLabel">d3</a></li><li class="vocabulary-links field-item odd" rel="dc:subject"><a href="/project7/?q=taxonomy/term/34" typeof="skos:Concept" property="rdfs:label skos:prefLabel">prediction</a></li></ul></span>Sun, 22 Dec 2013 07:04:17 +0000rakirk20 at http://clockworkchaos.com/project7http://clockworkchaos.com/project7/?q=nfl-yards#comments