For this I actually had sufficient data available that adding more didn't do much.
You need a testing dataset in order to validate the performance of the model. If you validate against the training set really what you're doing is measuring the model's ability to fit the training data - which it will be able to do with high accuracy. That will, however, result in a much diminished ability to predict any new games - as it isn't "learning" the features of college basketball as much as it is memorizing the contents of the training set.
As others pointed out, it'd probably even be better to add a third grouping which is tested against only after the algorithm has finished - as an objective validation against as of yet unseen data.
You need a testing dataset in order to validate the performance of the model. If you validate against the training set really what you're doing is measuring the model's ability to fit the training data - which it will be able to do with high accuracy. That will, however, result in a much diminished ability to predict any new games - as it isn't "learning" the features of college basketball as much as it is memorizing the contents of the training set.
As others pointed out, it'd probably even be better to add a third grouping which is tested against only after the algorithm has finished - as an objective validation against as of yet unseen data.