The goal of this project is to predict the outcome of soccer matches. In my opinion, this subject is quite original, and challenging. The resources that the team used were data sets containing the results of 25,000 matches including the formation of the teams, betting odds and information about individual players.
Talking about the technical details, I think that this team did a great job transforming the features of the data set (feature engineering), in particular, I liked how they transform the formation of the team and the particular skills of each player into grouped columns. I also liked that the team applied multiple techniques from the course starting from the simplest one to more complicated loss functions in their linear classification problems and also decision trees. Finally, I found particularly interesting that the group took several data sources and combined them to make a more robust model.
On the other hand, the first thing that raise a question in the work is why does the group deleted almost 20% of their data instead of trying other alternatives that we saw in class such as predicting the missing values or replacing them by the mean, etc. Besides that, in general this group could have done a slightly better job testing and reporting the results, I couldn’t find what kind of cross validation they used and it hard to see the results at the same time because there is not a table to summarizes everything. Finally, in this setting is probably not really important but the report could have been more formal (although I enjoyed the gifs).
The goal of this project is to predict the outcome of soccer matches. In my opinion, this subject is quite original, and challenging. The resources that the team used were data sets containing the results of 25,000 matches including the formation of the teams, betting odds and information about individual players.
Talking about the technical details, I think that this team did a great job transforming the features of the data set (feature engineering), in particular, I liked how they transform the formation of the team and the particular skills of each player into grouped columns. I also liked that the team applied multiple techniques from the course starting from the simplest one to more complicated loss functions in their linear classification problems and also decision trees. Finally, I found particularly interesting that the group took several data sources and combined them to make a more robust model.
On the other hand, the first thing that raise a question in the work is why does the group deleted almost 20% of their data instead of trying other alternatives that we saw in class such as predicting the missing values or replacing them by the mean, etc. Besides that, in general this group could have done a slightly better job testing and reporting the results, I couldn’t find what kind of cross validation they used and it hard to see the results at the same time because there is not a table to summarizes everything. Finally, in this setting is probably not really important but the report could have been more formal (although I enjoyed the gifs).