Multilingual Artificial Intelligence for Classifying Ethiopian FinTech User Reviews (Research Square)

Summary

• The initial dataset collection gathered 8,387 records from 27 Ethiopian FinTech applications, yielding 7,148 records after quality control procedures.p.2
• AfroXLMR attained the best overall performance among all models, achieving an accuracy of 0.967 and a macro-F1 score of 0.960.p.17

Key insights

1. AfroXLMR achieved the highest overall macro-F1 score of 0.960 and accuracy of 0.967 on the held-out test set.p.17