Data-Driven and Explainable Discrete Optimization for Effective Transportation in Healthcare

This project aims to contribute to the development of safe and trusted, artificially intelligent transportation in healthcare. The London Ambulance Service (LAS) operates more than 1100 ambulances to respond to medical emergencies in the London area. On an annual basis, LAS serves more than 1.2 million incidents. Effective ambulance positioning and allocation to incidents are critical decisions, while also accounting for the imprecise, annually increasing, and time-varying demand. Optimizing the incident response times with a clear human-trusted rationale behind the decision-making process poses a real challenge.

To this end, [Letsios et al. 2019, Liaskas 2020, Liu 2020] have developed discrete optimization methods for resource allocation problems arising in transportation companies, including LAS. These methods are used for descriptive analytics purposes, by identifying improvements in historically implemented solutions to these problems and providing useful, data-driven insights for future decisions. Typically, the allocation of ambulances to incidents is decided using the incidence criticality, location, and ambulance availability. Optimization software offers major benefits in transportation companies, but cannot be fully trusted by healthcare practitioners due to the lack of explainability. Argumentation-based models have been shown to be able to support decision-making in a transparent way. [Cyras et al. 2019, Cocarascu et al. 2020] proposed argumentation-based approaches that enable interaction with and generation of user explanations for decisions produced by optimization solvers and predictive analytics tools.

This project aims to enhance descriptive analytics with discrete optimization by developing robust integer optimization models and algorithms for ambulance management, taking into account the uncertainty and temporal nature of medical emergencies. It also aims to develop novel argumentation-based approaches for generating user explanations tailored to ambulance management.

[Cyras et al. 2019] Kristijonas Cyras, Dimitrios Letsios, Ruth Misener, Francesca Toni. Argumentation for Explainable Scheduling. p. 2752-2759, AAAI 2019.

[Cocarascu et al. 2020] Oana Cocarascu, Andria Stylianou, Kristijonas Cyras, Francesca Toni, Data-Empowered Argumentation for Dialectically Explainable Predictions. ECAI 2020, p. 2449-2456.

[Letsios et al. 2019] Dimitrios Letsios, Jeremy T. Bradley, Suraj G, Ruth Misener, Natasha Page. Approximate and Robust Bounded Job Start Scheduling for Royal Mail Delivery Offices. 2019.

[Liaskas 2020] Athanasios Liaskas. Data Analysis with Mixed-Integer Optimization for Royal Mail Centers. M.Sc. Thesis, Imperial College London, 2020.

[Liu 2020] Kaiyue Liu. Data Analysis for Deciding Ambulance Locations and Allocations to Incidents in London Ambulance Service. M.Sc. Thesis, King’s College London, 2020.

Project ID

STAI-CDT-2021-KCL-3

Supervisor