I have posted over the course of the last 6-7 years and have spent time predicting games with my computer model to a minimal success. It is very hard to find a system that can sustain a reasonable outcome.
So what I have changed---I am now running a coorelation analysis on zscores of individual statistics. What I mean is that I am looking at each statistic relative to the mean and their peers and then I am trying to determine whether the outcome has any coorelation. A coorelation score ranges from -1 to 1. In college the best I have gotten is around .56 on NFL I have only been able to generate a coorelation to about .38.
My next step was to take this coorelation and check it against recent data. The combination of stats that met my specs I then weighted to get optimal results. I put 2010-2014 in one category and the last two years in a second category. When I ran the analysis... I tested to see what combination of the weighted statistics produced a 58% outcome in each of the categories at the same time.
If it did...I kicked that combination to a new spreadsheet and I am now running those combinations to see if they work this year. I have different combinations for over, under, home, away, favorites and dogs. I chose to break them down into six different categories.
Week 1--Hit 58%--but I could only use last years end of season data
Week 2 Hit 51% but that was based on only one games stats
Week 3 Hit 46.3% but again these stats only include two volatile weeks for college football.
I start my betting Week 4 for College and Week 3 for Pro.
Any suggestions on what I could do better is appreciated. I bought computer parts to build a new computer because my 8 year old computer couldn't handle the calculations. Parts just came in...running NFL program right now... takes 6-8 hours...college about the same... I will post my picks so that the board can rip them apart.
Thank You
Dave