Incremental Ensemble of One Class Classifier for Data Streams with Concept Drift Adaption

Due to the digital era, and recent development in software and hardware technology uses enormous applications like e-commerce, mailing system, social media, fraud detection, weather and network application. These applications generate a huge amount of continuous, sequenced, temporarily ordered and infinite data called as a data stream. There is a need to manage such data streams with real-time responses and sufficient memory requirements. Data streams lead to a problem of changing data distribution of the target variable is called as the concept drift. The Learning model performance degrades if the concept drift is not addressed, so there is a need for a learning model that adapts the concept drift by retaining the good performance of the model. One-class classification is a promising research area in the field of data streams classification. In the One-class classification, only the positive samples are considered to address the class imbalance and drift detection problem by not considering their counterparts. In this paper, an Incremental One-class Ensemble classifier is used to adapt the concept drift problem in streaming data. Model is evaluated with the Spam and Electricity real-world datasets and the model is used to address Gradual and sudden drift with 82.30% and 81.50% accuracy.