In our analysis, there were several things we did not do, even though we had the option to. One of the things we decided against implementing was a form of  PCA dimensionality reduction. Although we intended to do this at the outset, we opted not to because we had no problem with the runtimes for any of these algorithms, and we did not have too many features to begin with. Also, we tried several different ways to do an iteration analysis - that is, check performance of the algorithms in relation to the number of iterations. However, both of the models we used set the number of iterations automatically, and only had a score once the model was finished fitting. Thus, we employed different techniques to ensure that the algorithms were converging, and spent more time looking at the other parameters.

 

In conclusion, we were satisfied with our results. Our ultimate goal was to get close to, or beat, the bookmakers prediction accuracy of 53%. While we never achieved a value higher than this, we were consistently getting results in the 51-53% range. Had we been given a more complete data set and more features regarding each team, 53% potentially could have been broken. However, we believe we were very successful given the data provided.