This study aimed to quantify the combined influence of traffic activity and meteorological conditions on roadside PM10 in Makassar, Indonesia, and to evaluate whether an integrated model provides better predictive performance than traffic-only and nonlinear alternatives. Field measurements were conducted at 30 roadside points between 25 February and 20 March 2025. PM10 concentrations were measured using a High Volume Air Sampler, traffic was represented by total vehicle volume, and temperature, relative humidity, air pressure, wind direction, and wind velocity were recorded concurrently. The analysis included the Shapiro–Wilk test, Pearson correlation, simple linear regression, multiple linear regression, fifth-order polynomial regression, Random Forest benchmarking, leave-one-out cross-validation, residual diagnostics, Moran’s I, and exploratory inverse distance weighting. Total vehicle volume was positively associated with PM10 (r = 0.752, p < 0.001), while relative humidity showed a strong negative association (r = −0.740, p < 0.001). The final multiple linear regression model using total vehicle volume and relative humidity achieved R² = 0.720 and adjusted R² = 0.699. Under leave-one-out cross-validation, it outperformed the other models with RMSE = 24.79 µg/m³, MAE = 18.62 µg/m³, cross-validated R² = 0.650, and observed–predicted r = 0.807. Random Forest ranked second but did not surpass the multiple linear model. No significant global spatial autocorrelation was detected in measured PM10 or model residuals, and IDW showed weak out-of-location predictive skill, so the spatial surface was retained only as an exploratory visualization. The study is limited by 30 point-level observations, selected one-hour monitoring periods, and the absence of external seasonal validation. Practically, the model provides an interpretable screening framework for identifying roadside locations where traffic intensity and humidity jointly indicate elevated PM10. Its originality lies in combining traffic, meteorology, internal validation, robustness diagnostics, a machine-learning benchmark, and explicitly constrained spatial interpretation within a single roadside assessment framework for Makassar.