Real-Time Anomaly Detection in IoT Sensor Data Using Statistical and Machine Learning Methods

Abstract

The proliferation of the Internet of Things (IoT) has intensified the need for robust real-time anomaly detection to safeguard system reliability and operational efficiency. This study presents a comprehensive framework that integrates statistical methods with machine learning techniques for anomaly detection in IoT sensor data. Initial analyses employed descriptive statistics, correlation structures, Z-score, and Interquartile Range (IQR) to characterize data distributions, establish anomaly thresholds, and assess sensor stability. Results identified loudness as the most volatile feature, while light demonstrated high consistency. To enhance detection accuracy, multiple machine learning models were evaluated, including Logistic Regression, Decision Tree, Random Forest, Support Vector Machine (SVM), and Neural Network (MLP). Comparative findings revealed that ensemble and deep learning methods significantly outperformed traditional approaches, with the Neural Network achieving superior accuracy and Random Forest providing strong efficiency. The results highlight the potential of hybrid statistical–machine learning frameworks for effective, scalable, and interpretable real-time anomaly detection in IoT environments.