<p>Air pollution poses a significant threat to human health, yet traditional monitoring platforms and reference-grade instruments are costly and offer limited spatial coverage. This study developed a custom-designed sensor system that integrates electrochemical sensors with a microcontroller for data collection and processing. Various algorithms were tested for sensor data calibration, with the random forest algorithm delivering the highest accuracy and lowest error. Further analysis emphasized the importance of temperature and relative humidity in improving the predictive accuracy of the models, particularly for the calibration of CO and O_<i>x</i> sensors. Model performance was significantly enhanced by using larger training datasets, requiring 28 days of data for optimal calibration of CO and NO sensors. In contrast, the NO₂ and O_<i>x</i> sensors performed well with smaller datasets (7 days or more). The sensor system was further employed to monitor variations in particulate matter concentrations and examine the relationships among different pollutants. Results show that reducing NO_<i>x</i> and O₃ concentrations to 50% of their original levels led to a 25% and 5% decrease in PM, respectively, but their simultaneous reduction resulted in only a 15% decrease, suggesting a nonlinear interaction between these pollutants and PM. The system was also used to monitor air quality inside and outside subway stations, providing real-world validation of the performance of the sensor system and demonstrating its versatility. Overall, this study demonstrates the feasibility and effectiveness of the custom-designed sensor system for air quality monitoring, emphasizing its potential for high-resolution sensor network deployment and its capability to generate valuable data to inform urban air quality policy decisions.</p> Graphical abstract <p></p>