Effective CRM using Predictive Analytics

4 Classification algorithms

4.1 Data mining algorithms for classification

This chapter is dedicated to classification algorithms. Specifically, we’ll present Decision Tree, Bayesian network, and Support Vector Machine (SVM) models. The goal of these models is the same: to classify instances of an independent dataset based on patterns identified in the training dataset. However, they work in different ways. Bayesian networks are probabilistic models which visually represent the dependencies among attributes. Decision Trees perform a type of supervised segmentation by recursively partitioning the training dataset and identifying “pure” segments in respect to the target attribute, ideally with all instances landing on the same class. SVM models are machine learning models which use nonlinear mappings for classification.

Despite their different mechanisms, these algorithms perform well in most situations. But they also differ in terms of speed and “greediness” with Decision Trees being the lightest and fastest as opposed to SVMs which typically take longer training time and are also quite demanding in terms of hardware resources. Although we’ll try to present their pros and cons later in this chapter, the choice of the algorithm to use depends on the specific situation, and it is typically a trial-and-error procedure. In this chapter, we’ll focus on the explanation of the algorithms. But remember model training is just a step of the overall classification process. This process ...