Cholera Epidemic Prediction Using Internet News Data in West Africa: A NLP and Machine Learning Approach with Transformer-Based

Domaine:

healthcarenatural language processing

Type de record:

paper

Créateur:

K-KA.JB. O.O

Éditeur:

Cre

Hôte:

Cholera remains a persistent public health threat in West Africa, where formal surveillance pipelines are often delayed by weeks or months. This study introduces a Natural Language Processing (NLP) framework to classify cholera outbreak related news articles in HealthMap corpus. Contextual BERT encoder is combined with three downstream classifiers: Random Forest (RF), LSTM, and Bidirectional LSTM (Bi-LSTM). The framework is evaluated on 942 HealthMap articles for three (3) years using a temporally strict split. Experiments are repeated across five random seeds: A BERT + Bi-LSTM) achieves F1 = 0.750 ± 0.038 and PR-AUC = 0.817 ± 0.025 while fine-tuned BERT + Linear Head baseline reaches F1 = 0.778 ± 0.019, the highest overall F1. This indicates that the recurrent layer of features representation does not surpass the performance of end-to-end fine-tuning. The two models are statistically indistinguishable under McNemar’s test. McNemar’s tests confirm statistically significant improvements over TF-IDF + Logistic Regression (p = 0.003), Random Forest (p < 0.001), and LSTM (p < 0.001) baselines. The analysis shows BERT + Bi-LSTM achieves the lowest score (0.108 ± 0.012) due to small number of HealthMap news. The TF-IDF + Logistic Regression baseline achieves an F1 score of 0.583 in the temporal evaluation. In 5-fold cross-validation, it reaches F1 = 0.898 ± 0.013, underscore the critical importance of temporal evaluation for assessment. Keywords: Cholera surveillance; NLP; BERT, HealthMap, Temporal evaluation Proceedings Citation Format Abdullah, K-K. A., Ajayi, A.J., Efuwape, B. T., Solanke, O.O., Ologunleko, E.I, & Oladiran, O.E (2025): Cholera Epidemic Prediction Using Internet News Data in West Africa: A NLP and Machine Learning Approach with Transformer-Based. Proceedings of the 39th iSTEAMS Multidisciplinary Bespoke Conference & 39th Extended Conference 17th – 19th July/17th – 19th October, 2025. University of Ghana, Accra, Ghana. Pp 334-344. isteams.net dx.doi.org

Visit

doi.org

Tasks

text classification

Languages

Similaires

Data-Driven Machine Learning Techniques for the Prediction of Cholera Outbreak in West AfricaNews Text Classification for Wolaita Language Using Transformer-based Model and Deep Learning ApproachLEVERAGING MACHINE LEARNING FOR EARLY DETECTION AND PREDICTION OF CHOLERA OUTBREAKS IN NIGERIA: A DATA-DRIVEN APPROACHVisit-level prediction of missed HIV appointments using machine learning and transformer-based models in UgandaFlood Prediction in Africa: A Machine Learning ApproachEvaluation and Prediction of Covid-19 Disease Spread across Sixteen West Africa Countries Using a Machine Learning Approach

Data-Driven Machine Learning Techniques for the Prediction of Cholera Outbreak in West Africa

The cholera epidemic remains a public threat throughout history, affecting vulnerable populations li

News Text Classification for Wolaita Language Using Transformer-based Model and Deep Learning Approach

LEVERAGING MACHINE LEARNING FOR EARLY DETECTION AND PREDICTION OF CHOLERA OUTBREAKS IN NIGERIA: A DATA-DRIVEN APPROACH

Cholera remains a significant public health challenge in Nigeria, causing numerous fatalities annual

Visit-level prediction of missed HIV appointments using machine learning and transformer-based models in Uganda

Flood Prediction in Africa: A Machine Learning Approach

Evaluation and Prediction of Covid-19 Disease Spread across Sixteen West Africa Countries Using a Machine Learning Approach

The covid-19 pandemic disease created a great scare and massive economic, social, political, and int