A CLASSIFICATION MODEL FOR SENTIMENT ANALYSIS OF DEPRESSION USING NIGERIAN TWEETS

Abstract

Depression represents a significant public health challenge in Nigeria, yet detection tools lack cultural sensitivity to the country's unique linguistic landscape. This study developed and evaluated a classification model for sentiment analysis of depression using Nigeria-specific Twitter data, addressing the gap in computational psychiatry tools tailored to African contexts. The methodology employed Twitter API to collect 52,681 tweets containing depression-related keywords in Standard English, Nigerian Pidgin, and local expressions. After preprocessing with text cleaning, tokenization, and TF-IDF vectorization, five machine learning algorithms were implemented: Logistic Regression, Random Forest, Naïve Bayes, Support Vector Machine, and Multilayer Perceptron (MLP). SMOTE was applied to address class imbalance. Performance evaluation revealed the MLP classifier as superior with 89.7% accuracy and AUC 0.96, followed closely by Random Forest (88.9% accuracy, AUC 0.96), while traditional models demonstrated moderate effectiveness. The significant performance disparity between advanced models and conventional classifiers confirms the necessity of sophisticated computational approaches for detecting mental health indicators in Nigeria's linguistically complex digital discourse. These findings offer promising pathways for implementing scalable mental health monitoring. The culturally sensitive classification model developed provides a foundation for early intervention through automated monitoring of public social media discourse, potentially transforming mental health surveillance in Nigeria.