Modal emission models describe vehicular emissions with stratified vehicle kinetic conditions or engine parameters. They are widely adopted in regulatory applications in North America and Europe to estimate emissions and energy consumption from mobile sources. However, challenges exist in the development of modal bins, especially that previous approaches rely on manual adjustment and tuning, which increase the propensity to emission misclassification. This study proposes a new approach to generate modal bins, which overcomes the limitations of previous studies. It uses a Greedy algorithm to define optimal mode boundaries and improve model robustness. The model is calibrated with emission data from a portable emission monitoring system and validated against an independent dataset. Our modelling approach can effectively reflect carbon dioxide (CO₂) emissions in steady and aggressive driving conditions with errors lower than 7% at the trip level. The introduction of engine parameters is found to improve model prediction for carbon monoxide (CO) and nitrogen oxides (NO_x) by about 30% compared with the models relying on external variables.