Classical models classify Ghana mobile-money scam SMS on a 220-message set
Original title: Machine Learning-Based Detection of Mobile Money Scan Messages in Ghana: A Comparative Evaluation of Naive Bayes, Support Vector Machine, and Randon Forest Classifiers
Held-out accuracy, F1 and AUC were 1.00 for Naive Bayes and linear SVM; cross-validated accuracy averaged 92.0% to 96.6%.
Researchers trained Multinomial Naive Bayes, a linear SVM and Random Forest to separate scam and legitimate Ghanaian mobile-money SMS. The labelled corpus has 220 messages in 18 subcategories, with TF-IDF unigrams and bigrams after normalising amounts, phone numbers and USSD codes. The authors say the classifiers fit low-resource on-device filtering but warn that a small template corpus can inflate scores.
Unconfirmed: Perfect held-out scores are on a small template corpus, and the authors flag ceiling effects.
- Graphic Online figures cited: GHC518.4 billion in December 2025 and over 26.7 million accounts.
- Cross-validated accuracy averaged 96.6%, 94.9% and 92.0% for the three models.
- Scam n-grams included send, claim, PIN and pincode.
- Evaluation used stratified five-fold cross-validation and a 20% held-out set.
From the original article
1 min read at arxiv.orgMobile money (MoMo) is Ghana's preferred retail payment channel, recording GHC518. 4 billion in transactions in December 2025 and over 26. 7 million active accounts (Graphic Online, 2026).
By Ibrahim Salifu · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.