The application uses weather parameters put in by the user to return a decision whether or not it's a good idea to play tennis. The percentage displays the confidence in that decision based on the data set taught to the Naive Bayes. (Refinement of this prior data set is what makes it 'supervised' learning).