Predicting the Relationship Between the Size of Training Sample and the Predictive Power of Classifiers

The main objective of this paper is to investigate the relationship between the size of training sample and the predictive power of well-known classification techniques. We first display this relationship using the results of some empirical studies and then propose a general mathematical model which can explain this relationship. Next, we validate this model on some real data sets and found that the model provides a good fit to the data. This model also allow a more objective determination of optimum training sample size in contrast to current training sample size selection approaches which tend to be ad hoc or subjective.