CS-IBC: Cuckoo search based incremental binary classifier for data streams

Abstract

The act of classifying data streams is widely studied in the literature over the last decade. Incremental or progressive learning strategies are adapted to classify the data streams by many research contributions in recent literature. The contemporary affirmation of recent literature indicate that issues like timeliness, linearity of computational complexity, incremental update of the classifier, and concept drift adaptation in data stream classification are still significant constraints. And there is a need for an algorithm to provide good classification performance with a reasonable response time and maximal classification accuracy. In order to arrive at this, Cuckoo Search Based Incremental Binary Classifier (CS-IBC) has been devised in this manuscript. The contributions of the CS-IBC is to define class labels from training data and fasten the class search through bio inspired strategy called “CUCKOO Search”. A periodical update of the classifier is also proposed to update the classifier if a set of new labelled records are given. The CS-IBC is tested on KDDCUP data that contains records, which are labelled as attack prone or normal. Metrics such as classification error rate, latency of the classification strategy and classification accuracy deterioration were assessed to estimate the scope of the CS-IBC as binary classifier. The experimental study indicates that the proposed CS-IBC is robust and scalable.