Summary

Comparing the RMSEs produced by each of our models, we see that the support vector machine method results the lowest RMSE.

• Our best model RMSE was 0.35, which means our best model are on average 0.35 points off from the true rating. Not bad! Though there is definitely room for improvement, potentially with more variables.
• There are several board game categories and mechanics that tend to do better than others. To maximize the chances of a board game becoming popular, a game designer could attempt making a game in these top categories (such as storytelling games or war games) or using some popular mechanics (such as line drawing and deck/pool building).
• Even though our dataset is not super accurate at predicting rating, it can still be of great utility for building a board game recommender from its covariates! (This is what we will try for the final part of our project, to be continued).

• Limited descriptive characteristics of board games in our dataset.
• No pricing information, though may not have been that helpful.
• A lot of inherent lack of predictability in board game popularity, depends on many aspects, such as marketing, aesthetics, fluctuating public preferences, legacy factor, etc.

• To more accurately model board game popularity, we may consider a network model to examine the spread of board games amongst groups of friends and family. As most board games tend to be a social endeavor (75% of board games are multiplayer or party games in our dataset), it is perceivable that board games may become popular through personal contact and word of mouth.
• We can try to incorporate more features of the board games such as year of publication, minimum players required in the future to see if that gives better recommendations.
• We can also acquire some user data in future studies to make more personalized recommendations.