Abstract
Antibiotics are a distinct class of drugs that underpin contemporary medicine. In Nigeria, antibiotics is a hugely prescribed medication with up to 71.1% of antibiotics prescribed per encounter. Its massive use and importance mean that it is usually targeted for counterfeiting and there is the need to evaluate the prevalence of counterfeited antibiotics in Nigeria. This work used a machine learning model to evaluate the incidence of counterfeit antibiotic in the six geopolitical regions of Nigeria. Counterfeiting status of 764 randomly sampled antibiotics from all regions was obtained and separated into training and testing sets. Two versions of training data were generated with SMOTE resampling technique, the training data contain 16.4% counterfeited antibiotics while the two versions generated contain 40% and 50% respectively. Three binary logistic regression models B1, B2 and B3 were fitted to the training data and it two versions. The performance of the fitted models was assessed with relevant metrics and the Receiver Operating Characteristic (ROC) curve. The results disclosed a higher rate of counterfeit antibiotic in the three northern regions of Nigeria. The results also revealed model accuracies for B1=84.1%, B2= B3=69.6%. and model sensitivity value B1= 0%, B2 = B3=75.8%. The Area Under Curve (AUC) ROC scores of 0.63 and F2-score of 0.59 shows the inadequacy of the model to correctly predict counterfeited antibiotics. The work however revealed that the northern regions are more targeted for antibiotics counterfeiting than southern regions, suggesting there is a clustered spatial distribution of counterfeited antibiotics in Nigeria.
Keywords: Antibiotics, Geographical Regions, Model Accuracy Sensitivity, Specificity.