Advanced Modeling of Fuel Efficiency in Light-Duty Vehicles Using Gamma Regression with Log-Link Under Real Driving Conditions at High Altitude: Quito, Ecuador Case Study

Fuel efficiency (FE) modeling under real-world conditions remains limited in Andean cities, where topographical and traffic conditions affect vehicle performance. Vehicles powered by spark-ignition engines are the most popular in Latin America, but few studies integrate dynamic conditions with geographic features. This study addresses this gap by developing an explanatory model to predict FE for light-duty vehicles (LDVs) in the Metropolitan District of Quito (DMQ), which is one of the most congested cities in Latin America. Data were collected from eight vehicles circulating under real conditions across 35 zones in the DMQ. Predictors such as vehicle speed (VSS), acceleration (A), speed per acceleration in its 95th percentile (VA[95]), road slope, and Vehicle-Specific Power (VSP) were included in the analysis. As a first attempt, linear models were tested, but the assumptions were not satisfied. Therefore, a Gamma regression model with a logarithmic link was selected. The final model achieved a Root Mean Square Error (RMSE) of 0.939, a Relative RMSE (RRMSE) of 0.155, a Mean Absolute Error (MAE) of 0.754, and an approximate coefficient of determination (R2) of 0.956. This methodology combines continuous and categorical variables and offers a replicable framework for FE estimation in other urban contexts.