Page 433 - AI for Good Innovate for Impact

AI for Good Innovate for Impact

After the registration and recording of the applicant’s biodata, the RSD process under UNHCR’s
mandate includes three key stages:

• interviewing stage, where information is collected from the Applicant on their refugee
claim, 4.4-Productivity
• assessment stage, where eligibility officers make a legal analysis to determine whether
the applicant meets the refugee definition under the 1951 Convention and its protocol,
and
• reviewing stage, where the legal analysis is reviewed by a supervisor and either endorsed
or returned for re-interview or redraft of the legal analysis.
Despite efforts to ensure a timely issuance of decisions through the implementation of varying
processing modalities and strategies and the application of procedural standards for mandate
RSD, the duration of the process can range from a day to many months.

Lengthy RSD processing times pose significant challenges, particularly for Asylum-Seekers
awaiting decisions on their applications. The duration of these procedures can significantly
impact the well-being and safety of Asylum-Seekers. Uncertain refugee and resulting legal
status can hinder access to essential services, expose individuals to human rights breaches,
and delay access to long-term solutions.

Effective and timely decisions on refugee claims enable recognised refugees to find protection
and pursue socio-economic inclusion in their country of asylum. Additionally, prompt decision-
making supports the safe, dignified, and rights-based return of individuals who do not qualify
for international protection, reducing the misuse of asylum procedures and ensuring a better
use of scarce resources.

The high number of asylum applications made with UNHCR coupled with dwindling resource,
has led UNHCR to turn to data science to make an in-depth analysis of its RSD process and
leverage new technologies, including process mining and ML techniques, to identify both
bottlenecks and efficiencies in the RSD process.

With advanced data modelling and AI-powered tools, UNHCR aims to uncover root causes of
delays at each of the stages in the RSD process (interviewing, assessment and reviewing) under
its mandate, and devise ways to tackle such delays or prevent them altogether. This analysis
will also allow the identification of efficiencies in the RSD procedure, which could be replicated
and streamlined across UNHCR RSD procedures.

By combining advanced ML techniques with AI-driven process mining, the technological
approach identifies causes of delays in UNHCR’s RSD process. This integrated approach
leverages predictive models - such as linear regression, decision trees, and random forests—to
identify key drivers of processing delays, while also using process mining to visualise real-time
workflows and detect inefficiencies or bottlenecks in the system. The models are fine-tuned
through cross-validation and hyperparameter optimisation to enhance accuracy and reliability.
By uniting predictive analytics with dynamic process mapping, this solution offers a powerful,
data-driven framework for improving operational efficiency and guiding evidence-based
decision-making in refugee case management. Process mining can be viewed as a post-facto
analytical approach that examines event log data to uncover actual process flows, bottlenecks,
and variations within a system. In the context of the RSD process, it enables the identification
of patterns in the duration and progression of individual cases through various stages. By
reconstructing the real pathways followed, rather than relying solely on formal procedures or

397