Researchers Use ML to Detect Obfuscated Swahili Abuse, Focus on 100M+ Speakers

Researchers have developed machine learning models to detect obfuscated abusive language in Swahili, a low-resource language with over 100 million speakers, focusing on child safety. Using SVM, Logistic Regression, and Decision Trees optimized with SMOTE, the study addresses a critical gap in online safety for East Africa. While promising, the models' generalizability is limited by small, imbalanced datasets. The work advocates for expanded data and advanced techniques to improve cyberbullying detection systems.