Predicting factors in financial loss among Malaysian scam victims using machine learning

This study compares decision tree and logistic regression models to predict financial losses among 394 Malaysian scam victims and to identify key predictors. Model accuracy refers to overall classification accuracy on the validation set, alongside AUC, sensitivity, specificity, and F1 score. Model interpretability was operationalized as the transparency of decision rules in the fitted tree, that is, the ease of tracing split conditions along each decision path. On validation, the decision tree achieved higher AUC and accuracy than logistic regression (AUC 0.838 versus 0.797; accuracy 83.05 percent versus 72.88 percent), with gains in sensitivity and F1 score as well. Emotional harm emerged as the strongest predictor of financial loss, followed by cybersecurity knowledge and age, whereas gender, urbanity, education, and internet use contributed modestly. These findings support targeted victim support, stronger fraud detection, and population-level digital literacy initiatives.