Integrating Traditional Machine Learning Approaches with Explainable Anomaly Detection for Multimodal Sensor Data

Anomaly detection has recently become essential in the context of human activity recognition, especially with emerging technologies like the Internet of Things and smart environments. These technologies are key contributors to data streams, generating vast quantities of continuous data across various applications. Nevertheless, the ever-changing characteristics of these data streams present ongoing challenges that remain to be addressed to understand and interpret the predictions. Providing clear explanations for detected anomalies is crucial for trust and usability, especially in critical applications like healthcare. In an effort to address this challenge, we have employed various traditional machine-learning approaches such as SVM, KNN, DT, and RF alongside XAI techniques like LIME. This combination will be used to show the efficacy of traditional ML methods and to explain the prediction process of anomaly detection, offering insights into how these technologies can be effectively applied to multimodal sensor data. The results demonstrated that SVM, RF, and DT achieved an accuracy of 0.95, while KNN achieved an accuracy of 0.94. In the future, we aim to explore the integration of deep learning techniques with XAI to enhance the interpretability and transparency of human activity recognition models. By incorporating XAI methods, we seek to provide insights into the decision-making process of deep learning models, enabling users to understand the reasoning behind the detected anomalies and increasing trust in the system's outputs.