Budgeted Learning of Naive-Bayes Classifiers

There is almost always a cost associated with acquiring training data. We consider the sit­ uation where the learner, with a fixed budget, may 'purchase' data during training. In par­ ticular, we examine the case where observ­ ing the value of a feature of a training exam­ ple has an associated cost, and the total cost of all feature values acquired during train­ ing must remain less than this fixed budget. This paper compares methods for sequen­ tially choosing which feature value to pur­ chase next, given the budget and user's cur­ rent knowledge of Na'ive Bayes model param­ eters. Whereas active learning has tradition­ ally focused on myopic (greedy) approaches and uniform/round-robin policies for query selection, this paper shows that such methods are often suboptimal and presents a tractable method for incorporating knowledge of the budget in the information acquisition pro­ cess.