Building activities commonly generate substantial amounts of construction dust, adversely affecting the nearby environment and public health. Construction workers, in particular, face significant health hazards due to their prolonged exposure to elevated levels of this dust. Traditional method of monitoring individual exposure to construction dust, such as gravimetric samplers or high-end analytical instruments, are often expensive, cumbersome, and not suitable for real-time, widespread deployment. This study employs the low-cost sensors (PMS A003-G10) to measure dust concentrations in varied environments: first low, then high, and then once again low concentrations. In the first low-concentration environment, the G10 sensors showed strong correlation (R2 > 0.81) and acceptable error (RMSE<13.6 μg/m3). However, in high-concentration environment, the G10 sensor faced range limitation issues, yet maintained good correlation. Post high-concentration exposure, the G10 sensor exhibited increased NRMSE and MAPE, indicating adverse impacts on its measurement capability. To enhance the G10's performance in high concentrations, temperature and humidity were used as calibration factors. Four machine learning algorithms (MLR, RF, KNN, and XGBoost) were compared, with XGBoost demonstrating superior calibration (R2 > 0.96, RMSE<117.1 μg/m3). The model's generalizability was validated by integrating data from both low and high-concentration environments into the XGBoost training. Subsequent application to the second low-concentration dataset post high-concentration exposure assessed the model's generalizability and applicability. This study demonstrates that with appropriate calibration, low-cost sensors can effectively monitor individual exposure to construction dust across diverse concentration levels.