Sumobot is a prediction engine I wrote to help me select draft picks for our fantasy sumo league. Overall, its performance has been better than my own performance, but middling relative to the other members of my league.
Since we select our rikishi by snake draft, at every turn of the draft the ideal pick is the rikishi who is likely to win the most among the remaining rikishi. Thus, Sumobot gives a predicted ranking of rikishi from highest wins to lowest wins, and for each pick I select the highest rikishi from the prediction list that has not already been picked.
The prediction comes from analyzing historic data using Python and Pandas to determine a number of parameters which tend to correlate with higher scores. Those parameters are then scored against the available rikishi to generate a predicted ranking of rikishi by expected wins.
To judge how well the prediction did after the fact, Spearman’s rank correlation coefficient can be used to compare how close the predicted win ranking was to the the actual win ranking. This gives a score from -1 to 1, where 1 means the predicted rank order was exactly right, -1 means the predicted rank order was exactly backward, and zero is effectively no correlation (e.g., roughly the same as random guessing).