A Novel Direct Formula Based Technique for Predicting the Class Label of New Unseen Test Tuple without Learning

Supervised learning is an important task in many machine learning applications. In this paper, a direct formula is proposed for predicting the class label of a generalized new unseen instance. The mostimportant attractive feature of the proposed class label prediction formula is that it requires no training step. In other words, intermediate learning steps are not required and as a result of this there exists no intermediate models for class label prediction. Proposed formula directly predicts class label of thenew test tuple. Proposed class label prediction formula is based on the aggregate class label counts of the values of the attributes of the given entire training dataset. For optimal performance, proposed formula is normalized and normalized formula is used for prediction. For brevity, proposed class label prediction technique is named as Direct Class Label Prediction (DCLP). Experiments are conducted on many of the UCI machine learning datasets.