Utilizing Machine Learning to Mitigate the Impact of Absent Sensor Data on RTOS-Enabled Internet of Things (IoT) Systems

The Internet of Things (IoT) is increasingly reliant on Real-Time Operating Systems (RTOS) to enhance reliability. However, IoT systems suffer from reliability issues stemming from their inherently distributed architecture, particularly in regions with weak wireless network coverage, leading to the occurrence of missing sensor data. This absence of critical data can be attributed to sensors disconnecting sporadically, remaining offline, or shutting down due to various factors. Issues leading to data loss are unreliable network connectivity, limited power available through battery/solar, device heterogeneity, and others. These challenges may impede the seamless transmission of sensor data between sensors, IoT nodes, processing components, and the cloud. In high-stakes IoT systems, the absence of data can result in system unreliability or even failure, particularly when sequences of anticipated sensor values are absent. In response to these scenarios, this paper studies mechanisms for imputing missing sensor data, this way mitigating their impact on IoT systems. The proposed solutions utilize machine learning techniques, evaluated alongside statistical methodologies. These imputation strategies aspire to sustain the normal operational state of IoT systems, even when confronted with missing data, ultimately fortifying their reliability. Evaluation of several imputation methods is presented, including MICE, KNN, RF, Mean and Median imputation. Imputation based on RF consistently generated low error imputed values thus keeping the system in a stable state.