Driver Behaviour Based Assignment On Ride-Hailing Dispatch: A Gojek Use Case

Nusa, Septia (2026) Driver Behaviour Based Assignment On Ride-Hailing Dispatch: A Gojek Use Case. Masters thesis, Institut Teknologi Sepuluh Nopember.

Text 6025231012-Master_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (3MB) | Request a copy

Abstract

Ride-hailing platforms rely on real-time dispatching systems to assign drivers with customer requests, typically optimizing for revenue, pickup distance, and conversion rate (the ratio of fulfilled customer's requests to total requests created). While effective operationally, these objectives lack transparency from the driver’s perspective. In low-demand areas, strong driver effort does not always translate into higher earnings due to exogenous factors such as weak demand and spatial imbalance, leading to unclear prioritization and weak alignment between effort and earnings. This paper addresses this gap by incorporating driver-controllable factors into the dispatching objective. We implement a driver behaviour score derived from acceptance rate, completion rate, and online duration as a proxy for effort, and integrate it into a assignment model to strengthen the link between effort and earnings. However, directly enforcing this score introduces efficiency trade-offs, particularly longer pickup distances when high-scoring drivers are poorly positioned. To mitigate this, a spatial value component based on grid-based learning is introduced to estimate the future demand potential of origin and destination areas. The aligned seven-day experiment evaluates 78,981 orders per configuration. With grid learning disabled, increasing the behaviour weight, lambda, from 0 to 0.2 improves the Spearman score and income correlation from 0.807 to 0.877 and increases the conversion rate from 85.633 percent to 86.813 percent. However, the median pickup distance increases from 317.2 metres to 380.4 metres, while the Gini coefficient rises from 0.638 to 0.682. When grid learning is enabled at a lambda value of 0.2, the conversion rate increases further to 86.914 percent, the median pickup distance decreases to 377.3 metres, and the Gini coefficient decreases slightly to 0.680, while the Spearman correlation remains high at 0.871. Overall, the proposed approach provides a better balance between effort and income alignment and operational efficiency while maintaining an interpretable assignment mechanism.

Item Type:	Thesis (Masters)
Additional Information:	source: https://drive.google.com/drive/u/1/folders/1vJd9xaxYqp8Vopu0s810DTgDLjUXJn6Z
Uncontrolled Keywords:	ride-hailing, real-time dispatching, driver behaviour score, effort–income alignment, assignment optimization, grid-based learning, spatial efficiency, pickup distance
Subjects:	T Technology > T Technology (General) > T57.6 Operations research--Mathematics. Goal programming
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Informatics Engineering > 55101-(S2) Master Thesis
Depositing User:	Septianusa ,
Date Deposited:	03 Aug 2026 09:01
Last Modified:	03 Aug 2026 09:05
URI:	http://repository.its.ac.id/id/eprint/140971

Actions (login required)

View Item