Predictive diaspora protection: Evaluating AI driven early warning systems for threat mitigation among Nigerian migrants in South Africa

Abstract

The recurring outbreaks of xenophobic violence against Nigerian migrants in South Africa expose critical gaps in anticipatory intelligence and cross‑border protective coordination. Despite decades of diplomatic engagement between the Federal Republic of Nigeria and the Republic of South Africa, the institutional response to xenophobic violence has remained largely reactive, characterized by post incident condemnations and temporary evacuations rather than systematic threat anticipation. This study investigates the feasibility of Artificial Intelligence (AI) driven social listening tools to significantly predict pre‑violence digital escalation signals at least 48 hours before physical xenophobic attacks, and if the absence of a formalized interoperability framework between Nigerian and South African security institutions diminishes the translation of AI‑generated early warnings into effective protective interventions? The study adopts a quantitative predictive research design integrating multilingual Natural Language Processing (mBERT) classification of hate speech with logistic regression violence‑onset modelling across 120 observation periods covering the 2019 and 2022 xenophobic episodes. Model performance was evaluated using AUC‑ROC, F1 score, precision‑recall metrics, and out‑of‑sample temporal validation, while institutional moderation effects were assessed through interaction models and intervention‑success regressions. The findings demonstrate that AI‑derived indicators particularly mobilization signals and politically amplified hate speech significantly predict violence within a 48–72 hour window, achieving strong discriminatory accuracy (AUC > 0.85). However, the study also finds that early‑warning lead time translates into successful protective intervention only when supported by robust institutional interoperability between Nigerian and South African authorities. The study concludes that predictive intelligence is necessary but insufficient without structured coordination mechanisms. This study recommends the establishment of a 24/7 AI‑enabled “Xenophobia Watch” Joint Operations Centre as an institutional vehicle, staffed by NIA, NSCDC, NiDCOM Liaison officers with a secure Coordinating Channel to South African Police Service, translating AI generated intelligence into actionable protection responses, under the Nigeria–South Africa Bi‑National Commission and the institutionalization of a binding Diaspora Security and Interoperability Protocol to ensure that early‑warning signals consistently trigger timely diplomatic, law enforcement, and consular responses. Together, these measures advance a hybrid prevention model that combines technological foresight with institutional readiness.