A hybrid probabilistic and combinatorial optimization approach to analyzing rowing performance measures

Availability of small amount of data is a major barrier that challenges data mining methods in some applications. To overcome the low accuracy problem, due to system training, we propose a novel hybrid and deterministic approach and test it on rowing championship data. Rowing is a world championship and Olympic sport which requires strategic planning. For rowers to become medallists, they need to develop plans respective to each split (generally every 500 meters). We analyzed previous rowing competition data to find the optimal combination of the rankings and times at different splits that may result in the highest chance of finishing the race in certain places. We applied our novel hybrid and deterministic approach that, given the rankings and times at each split, employs a Bayesian Belief Network to find the likelihood of each final standing occurring. Our approach then utilizes the Tabu Search discrete optimization method to find a permutation of the rankings and times that maximizes the probability of finishing in a certain position. The results of our new hybrid methodology, given the small size of the available rowing data set, are more promising for strategic rowing planning than that of traditional data mining techniques e.g., state-of-the-art machine learning-based classification methods.