Domination Dependency Analysis of Sales Marketing Based on Multi-label Classification Using Label Ordering and Cycle Chain Classification
Boonyarit Soonsiripanichkul, Tomohiro Murata · 2016
Multi-label classification (MLC) is a technique that is used to solve conditional problems where the decisions are a set of labels. Classification is the learning task of using historical examples to make a model of conditions, so as to make decisions of unseen examples. To improve decisions in MLC, we can use the advantage of domination analysis or dependency relations between the classes by using an enhanced Bayesian chain classifier (BCC). We introduce an approach for chaining classifier primary order by its individual label accuracy priority (LPC-CC). Our method considers the dependencies among labels based on label accuracy priority ordering. Thus, binary relevance (BR) theory is used for label sequencing priority, and a cycle classifier chain using naive Bayes is used for finding domination. The model has been tested on two well-known benchmark datasets named Yeast and Emotions, and on a collection of car sale records from a Thailand automotive company.