Classifier learning with data sets comprising of imbalanced class distributions is a challenging problem in data mining. This has a significant impact on some of the real-world applications such as medical diagnosis, anomaly detection and fault diagnosis. This empirical investigation focus on two issues, i)Efficacy of machine learning deprived due to scarcity of observations in the class of interest ii) A vulnerability assessment in terms of privacy compromise when publishing health care data. The class balancing methods are analyzed for the suitability towards a suite of classification algorithms. The results indicate a positive synergy between the preprocessing method SMOTE- TL and the learning models. An identity management algorithm based on decomposition strategy is proposed for the class imbalance problem. This can have an impact on the healthcare industry where a published health record demands privacy without compromising machine learning results. © 2020 IEEE.