Predicting Fraud in Mobile Money Transactions using Machine Learning: The Effects of Sampling Techniques on the Imbalanced Dataset

Mobile Money Fraud is advancing in developing countries. We propose a solution to this problem based on machine learning. Labeled data from financial transactions which include mobile money are, however, skewed towards the negative class. Machine learning models built with such datasets are unreliable as prediction algorithms will be biased investigate performance of different sampling and weighting techniques Adaptive Synthetic Sampling (ADASYN) Minority Oversampling Technique (SMOTE). select Logistic Regression for experiments due its simplicity relatively low computational needs. The evaluated metrics. Manually tuning weights classes achieved best results our experiments.