TRAIN Scholar Steven Gonzalez's Blog

  Hi all, After making some final tweaks to our models to fix any errors, we went ahead and evaluated the accuracy of each model. The model that accounted for home-field advantage for its expected goal calculation ended up being far and away the most accurate out of the bunch. This model accurately predicted the outcome of 62% of the matches in last year's playoffs. All of the other models we evaluated ended up having less than forty percent accuracy. Pretty interesting to see just how big of an impact playing at home vs away has on a team's performance. So, with our model selected, we were ready to move on to the next phase. The playoffs actually start this weekend, so we quickly got to work on building a prediction for the first round of this year's playoffs. It should be fun to see how our model fares in its predictions this weekend.   

  Hi all, This week we finally had the whole team together for our weekly meeting and things went great. We went over our different models and we are now going to be comparing how they fared in predicting match outcomes when applied to last year’s playoffs. We did find some errors in our models, so we are now tidying things up to make our predictions for this year's playoffs. The regular season ends in a few days, so we want to have all our ducks in a row before the playoffs start. So for now, we are all working on making some final tweaks to our models and also going to be delving a bit deeper into binomial deviance and squared deviance to see if this will provide us the best barometer for error in our models.     

  This week I worked on a model to apply to last season's playoffs in order to see how its predictions compared to the actual outcomes of the first round of playoff play. The model that I was assigned used different parameters to determine expected goals (xG) for teams depending on whether they were hosting the match or were the visiting team. For the 6 home teams in the first round of the playoffs, the model only considered goals scored while playing at home throughout the regular season for its xG calculations. For the 6 visiting teams, the model only considered the goals scored in away games that season to calculate xG. This model accurately predicted the result for four of the six matches in the first round of the playoffs. It should be interesting to see how the models from the rest of the group performed. This week, Dr. Posta tasked me with an additional model that only takes the results of the last two games of the regular season for each of the 12 playoff teams in the f...