Text this: An Effective Data Sampling Procedure for Imbalanced Data Learning on Health Insurance Fraud Detection