A Machine Learning Model for Assessing the Prevalence of Counterfeit Antimalarial in Geographical Zones of Nigeria

Introduction: Antimalarial is listed among the most common type of live saving medicines that are counterfeited. In Nigeria counterfeited antimalarial continue to pose a great threat to the health of the citizens and there is the need to assess its incidence within the country’s six zones. This study assessed the prevalence of counterfeited antimalarial within the six geographical zones of Nigeria and the impact of zones on counterfeiting using a machine learning model for classification. Methodology: Secondary data on 2442 antimalarial collected from all the states in Nigeria were grouped based on geographical zones. The medicines were tested for originality using the gold approach for detection of counterfeit medicine; the Standard Scientific Laboratory (SSL) Data was separated to 70% training and 30% testing and 10- fold Cross Validation (CV) was performed. The training set was used to derive the models while the test set was used to evaluate the performance of the models. Three varieties of the training data were generated using the Synthetic Minority Oversampling Technique (SMOTE). This was done to ensure a more accurate prediction and a better model performance. Binary Logistic Regression (BLR) models were thereafter fitted to the training data and the three varieties of its resampling. The four models namely M1, M2, M3 and M4 were fitted with data containing 33%, 40%, 45% and 50% of the counterfeited antimalarial class respectively. The performance of the four fitted models were assessed with metrics like sensitivity, specificity and model accuracy. Results: The results showed that there is a higher incidence of counterfeited antimalarial in the north-east and south-east zones than in the other four zones of Nigeria. The work also revealed model accuracies of 67%, 65.8%, 65.8% and 56,8% for M1, M2, M3 and M4 respectively. M1 was biased as it did not correctly predict any counterfeited antimalarial. M2 and M3 performed better than M4 in terms of model accuracy and specificity while M4 performed better only in terms of model sensitivity. Conclusion: Overall, only 66% of antimalarial was correctly predicted by the best performing model. This suggest that zone alone is not adequate to classify or predict originality/counterfeiting status of Antimalarial in Nigeria.