{"title":"OASIS: Online adaptive ensembles for drift adaptation on evolving IoT data streams","authors":"T. Anithakumari, Sanket Mishra","doi":"10.1016/j.iot.2025.101545","DOIUrl":null,"url":null,"abstract":"<div><div>In this work, our proposed OASIS framework utilizes adaptive ensembles to accommodate IoT data drift. In this work, we introduce an innovative sliding window approach using periodograms, engineered to efficiently feed models with data input. Six distinct online learners, alongside three drift adaptation algorithms: EDDM, HDDM-A and ADWIN have been tested using various feature selection methods, such as particle swarm optimization (PSO), dragonfly optimization (DA), grey wolf optimization (GWO), genetic algorithm (GA), and whale optimization algorithm (WOA), which have been carried out to validate the efficacy of the OASIS framework. We introduce a weighted probability approach derived from multiclass outcomes to ascertain the most suitable learners for leverage bagging or voting ensemble application. This is followed by an optimal scoring mechanism to determine the best training set based on accuracy and execution time criteria. The selection of models is guided by a probability-based algorithm coupled with a scoring system. Furthermore, we benchmark three state-of-the-art drift adaptation frameworks to evaluate their performance relative to our proposed framework. Evaluations in the context of EDGE-IIoT demonstrated outstanding accuracies of 98.98% in binary scenarios and 99.92% in multiclass scenarios, with the IoTID20 datasets achieving notable accuracies of 99.94% in binary and 100% in multiclass scenarios, thus surpassing previous methodologies. The framework undergoes extensive experiments with two recent multiclass datasets, namely the Aalto and RT-IoT 2022 datasets, in which OASIS achieved 99.99% accuracy on the Aalto dataset and 96.52% on the RT-IoT 2022 dataset. Additionally, we compare our framework with various concept drift datasets and leading drift ensemble frameworks for performance comparison.</div></div>","PeriodicalId":29968,"journal":{"name":"Internet of Things","volume":"31 ","pages":"Article 101545"},"PeriodicalIF":6.0000,"publicationDate":"2025-02-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"Internet of Things","FirstCategoryId":"94","ListUrlMain":"https://www.sciencedirect.com/science/article/pii/S2542660525000587","RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q1","JCRName":"COMPUTER SCIENCE, INFORMATION SYSTEMS","Score":null,"Total":0}
引用次数: 0
Abstract
In this work, our proposed OASIS framework utilizes adaptive ensembles to accommodate IoT data drift. In this work, we introduce an innovative sliding window approach using periodograms, engineered to efficiently feed models with data input. Six distinct online learners, alongside three drift adaptation algorithms: EDDM, HDDM-A and ADWIN have been tested using various feature selection methods, such as particle swarm optimization (PSO), dragonfly optimization (DA), grey wolf optimization (GWO), genetic algorithm (GA), and whale optimization algorithm (WOA), which have been carried out to validate the efficacy of the OASIS framework. We introduce a weighted probability approach derived from multiclass outcomes to ascertain the most suitable learners for leverage bagging or voting ensemble application. This is followed by an optimal scoring mechanism to determine the best training set based on accuracy and execution time criteria. The selection of models is guided by a probability-based algorithm coupled with a scoring system. Furthermore, we benchmark three state-of-the-art drift adaptation frameworks to evaluate their performance relative to our proposed framework. Evaluations in the context of EDGE-IIoT demonstrated outstanding accuracies of 98.98% in binary scenarios and 99.92% in multiclass scenarios, with the IoTID20 datasets achieving notable accuracies of 99.94% in binary and 100% in multiclass scenarios, thus surpassing previous methodologies. The framework undergoes extensive experiments with two recent multiclass datasets, namely the Aalto and RT-IoT 2022 datasets, in which OASIS achieved 99.99% accuracy on the Aalto dataset and 96.52% on the RT-IoT 2022 dataset. Additionally, we compare our framework with various concept drift datasets and leading drift ensemble frameworks for performance comparison.
期刊介绍:
Internet of Things; Engineering Cyber Physical Human Systems is a comprehensive journal encouraging cross collaboration between researchers, engineers and practitioners in the field of IoT & Cyber Physical Human Systems. The journal offers a unique platform to exchange scientific information on the entire breadth of technology, science, and societal applications of the IoT.
The journal will place a high priority on timely publication, and provide a home for high quality.
Furthermore, IOT is interested in publishing topical Special Issues on any aspect of IOT.