Anomaly detection framework for IoT-enabled appliances using machine learning

doi:10.1007/s10586-024-04461-z

Anomaly detection framework for IoT-enabled appliances using machine learning

2024-04-30

1

PRE

AI

M

Mohd Ahsan Siddiqui *

M

Mala Kalra

DOI：10.1007/s10586-024-04461-z

Original

Original request for help

Save

Abstract

En 中文

Addressing the contemporary complexity inherent in anomaly detection within heterogeneous systems is paramount. This paper presents a novel methodology tailored to pinpoint appliance anomalies within the framework of Internet of Things (IoT) technology. By amalgamating the capabilities of IoT and Machine Learning (ML), this approach not only heightens the precision and dependability of anomaly detection but also serves as a practical solution; industrial applications. To better align with the current industrial landscape, we emphasize the practical implications of our work. Our methodology is designed to cater specifically to industrial needs, offering a solution that can be seamlessly integrated into existing systems, thereby enhancing operational efficiency and reliability. The core of our approach lies in employing a hybrid method, utilizing both the Facebook Prophet and Isolation Forest ML algorithms; robust and intelligent anomaly detection. This dual-layered strategy, integrating; ecasting and classification objectives, ensures a comprehensive approach to anomaly detection tailored; industrial settings. Evaluation of our methodology involves rigorous testing against real-time and emulated datasets, as well as comparison with existing methods. MSE has also been calculated using DT, RF, SVM, NB and Logistic supervised regressor. The Facebook Prophet model's accuracy, assessed using Root Mean Squared Error (RMSE), demonstrates its proficiency in forecasting values closely aligned with reference data points. Meanwhile, the IForest Unsupervised ML model excels in identifying anomalies, achieving high accuracy rates across various contamination levels. Through meticulous cross-validation, our proposed method exhibits significant accuracy, with rates of 93.60% and 95.72% on real-time and emulated datasets, respectively. The hybrid model (Fbprophet + iforest) has an average accuracy of 96.35% on (real-time + emulated). These results underscore the efficacy and reliability of our approach in industrial anomaly detection scenarios. In summary, by addressing the present-day challenges in anomaly detection and offering a tailored solution for industrial applications, our methodology stands poised to significantly enhance the efficiency and reliability of systems operating in diverse industrial contexts.

Keywords:

Anomaly detection system

Appliances' anomalies

IoT

IoT flock

Machine learning

Anomaly detection framework for IoT-enabled appliances using machine learning

Abstract

Journal

Organization

Cited Papers