Is customized bus service for commuter segments the need of the hour? An integrated IPA-machine learning framework to redefine commuter segments based on quality expectations

Bus service in emerging countries faces significant challenges such as inferior quality, limited appeal, and declining ridership, which is further aggravated with increased affordability of commuters. The conventional approach of providing a standardized service to all commuters has proven ineffective due to the diverse perception and requirements of commuters. Unlike traditional approaches that categorize commuters solely based on car ownership, the present study employs a comprehensive approach to segment commuters based on their service quality expectations and identifies improvement areas specific to each segment. The methodology uses integrated Importance Performance Analysis-Machine learning (IPA-ML) based framework to identify segments and improvement areas. While demonstrating the methodology in an Indian metro city, two segments, namely ‘tolerant users (low quality acceptors)’ and ‘quality seekers (high quality seekers)’, are identified using complete-linkage-hierarchical clustering with Naïve Bayes Machine Learning classifier. Intervention areas specific to these segments are identified using revised-IPA with Kernel Initialization technique giving due considerations to the segment wise factor structure and management schemes. The findings indicated the impact of multiple socio-economic factors on segment formation emphasizing the need for a holistic approach beyond car ownership when segmenting commuters. The significant differences in perceived improvement areas between the two segments highlight their distinct perceptions and expectations, indicating the necessity for segmenting the bus service to cater to their specific needs. While study provides case-specific findings for the improvement of bus service, the methodology and experience give a new direction to develop bus as a sustainable demand management instrument in a broader context