Crisis Informatics and Applied Social Media Mining: End-to-End Multi-Stage ML Pipeline

Abstract—This paper presents an end-to-end machine learning
approach for mining disaster-related social media content in
Ghana, applying text classification, credibility assessment, and
sentiment analysis to tweets. Using Support Vector Machines
and Naive Bayes with TF-IDF feature extraction and SMOTE for
class balancing, the study achieves robust accuracy and highlights
the prevalence of negative sentiment in crisis communications.
The results emphasise the critical role of data balancing
and feature engineering in extracting credible and actionable
information for disaster management. The findings contribute
practical methods for improving the response to real-time crisis in
West African contexts and suggest directions for future research
in multilingual and platform-inclusive analysis.
Index Terms—Crisis informatics, social media mining, disaster
prediction, credibility assessment, machine learning, sentiment
analysis