A Study on the Applicability of the Memory-Based Reasoning Classifier

펼치기

In recent year, many methods suitable for classification problems have been extended to include a range of popular techniques, such as neural networks, logistic regression and decision tree induction. Unlike other data mining techniques that use a training set of preclassified data to create a model and then discard the training set, for MBR (memorybased reasoning), the training set essentially is the model. This study gives a way on the memory-based reasoning, decision tree, logistic regression, neural networks and bagging model comparison methods for home equity lines of credit data using 1:1, 1:2, 1:3 and 1:4 target rate datamarts. Through the reasoning underlying their development, MBR classifier can also be a good choice to make a prediction. The proper k for MBR classifier is selected based on the minimum misclassification rate criterion. Under the proper k, we found that the performance of MBR dominated other classification technique for the data set that we analyzed.