Midterm Peer Review

[ngjulia]

Summary:
This project is using historical NFL data to find the best fantasy football point project model in a season given some starting lineup. The project uses data from two sources, which are joined together by a single game_id to get not only per player data but also game data as well. Feature engineering and preliminary models has been done that shows some promising data, but more regression and analysis is needed.

What I liked:

• I thought all your plots were really easy to read, and the analysis that supplemented the graphs was insightful -> the heatmap was particularly cool!
• I also liked how you distinguished the cumulative data and the past data, as I think that contextualizes a given player's performance really well... I'm excited to see how that works later on with other models
• I also think you guys highlighted well the places were you saw you needed improvement, and also already had included the use of rank deficient models to help with your error

Future Work:

• One thing I am a bit concerned about is the scale - in the plan you seem to want to incorporate more data, but I'm worried about how much more complexity that adds to the project -> you want to make sure you can properly clean the data and also use it effectively. I'd make sure to think about what you think this extra data can tell you and if it will meaningfully contribute to your problem statement
• Your exact problem statement is a bit unclear -> perhaps explaining exactly what your results are in a bit more clarity would help frame the report better. Is the point to get some point value from your model and then compare that ("backtest" in your report terms) with what actually happened?
• Since many of the features in the player dataset seemed to be tied to a particular position, is there a better way to model this data? As if someone is a running back, you might not care about a lot of the features and there could be some bias in your results. Also, I would maybe consider the different distributions of the players - I'd assuming there are less quarterbacks in the dataset than there are linemen, for instance, and consider how that could affect your results as well.

Otherwise, I think you guys have a good plan going forward and I am excited to see your results!