A Wireless Sensor Network LCS-AQMS: Design, Instrumentation, Calibration Modelling, and Data Analysis

The utilization of low-cost sensors (LCS) of air quality monitoring systems (AQMS) such as MQ7, GP2Y-Sharp, ZE15-Winsen, and ZH03 in air quality monitoring systems has seen a marked increase. This can be attributed to the cost advantages and ease of integration these sensors offer. However, the performance of these sensors is significantly impacted by environmental factors, including temperature and humidity, which result in substantial deviations from the reference values. The objective of this study is to develop a reliable AQMS system that applies a machine learning-based environmental compensation calibration model to improve the accuracy of CO and PM sensor readings, particularly in the campus environment of the Bandung Institute of Technology. A combination of CO and PM sensors is used to collect data, which is then temporally synchronized with data from a gravimetric reference sensor. This model was developed using a supervised learning approach with three main algorithms: linear regression, random forest, and gradient boosting. The input features encompassed temperature, humidity, the day of the week, and the raw PM sensor readings. The evaluation process employed hold-out validation to ensure the consistency of the model. The results of the study show that, despite its simplicity, linear regression provides the most stable performance, as indicated by the highest ${R}^{2}$ value and lowest error when used with limited data. This model is considered the most viable option for direct implementation on edge devices, due to its high computational efficiency and minimal resource requirements. Meanwhile, random forest or gradient boosting models are expected to yield better results when supplemented with training data such as wind direction, wind speed, and rainfall. In parallel, this study also developed a trustworthy wireless sensor network architecture that incorporates a Markov Chain-based reliability model to mitigate the risk of data transmission failure. This probabilistic approach enhances the robustness and continuity of environmental data collection in real-time deployments. The results of this study indicate that, with optimal model design parameters, LCS sensors can be reliably used for field measurements, thereby increasing the acceptability and credibility of the data. These results support the development of adaptive and energy-efficient air quality monitoring systems.