A Multinomial Logit Model for Urban Transportation Mode Choice in Nasiriyah City

Analyzing mode choice is essential for transportation planning, especially in urban areas where commuting significantly affects peak-hour congestion. The travel behavior of city residents regarding their transportation mode choices is crucial, as it impacts the overall efficiency of movement within the city. The multinomial logit (MNL) model is a valuable tool for estimating mode shares when travelers are presented with multiple travel options. This research investigates mode choice in Nasiriyah by applying the MNL model to identify factors that affect transportation preferences within the community. The model incorporates twelve explanatory variables that have not been previously examined in the urban context of Nasiriyah. These variables include income, age, household size, travel distance, travel time, travel cost, road conditions, availability of taxis, availability of public transportation, land use density, cultural characteristics, and safety concerns. The results revealed that private vehicle ownership is the preferred mode of transportation, closely followed by taxi services, indicating a strong inclination toward private vehicles due to their flexibility and convenience. In contrast, public transportation options, such as buses, ranked third due to inadequate service attractiveness stemming from limited geographic coverage and passenger comfort. Non-motorized modes, such as motorcycles and bicycles, received lower preference due to challenging weather conditions and insufficient infrastructure.  Additionally, the results indicate that the choice of transportation modes in Nasiriyah City is shaped by various factors, including economic conditions, travel impedances, infrastructure quality, service availability, land development patterns, and safety considerations. This underscores the immediate need to improve public transportation services, as well as infrastructure for pedestrians and cyclists. The model was trained and tested using specific datasets, ultimately achieving an F1-Score of 85.3% on the testing dataset.