Accurate Data Sampling Methods for Medical Data – Survey

In the internet age, data generation is rapidly increasing. Artificial intelligence, particularly big data classification technology, is becoming crucial in aiding medical auxiliary diagnosis research. However, medical big data often faces imbalances due to varying sample collections. Imbalanced data generally refers to a condition in which multiple samples of data within a given problem are not evenly distributed, thereby leading to underrepresenting one or more classes in the dataset. Conventional classification methods typically assume that each class has a comparable number of samples and that training misclassification costs are the same for all classes. However, compared to samples from healthy individuals, the misclassification cost of sick samples is higher. Therefore, it is a pressing issue to figure out how to improve patient identification without degrading the classification of healthy people. The study experimented with different SMOTE oversampling and under-sampling methods. The purpose of this work is to balance the imbalanced data using different sampling techniques. Two medical datasets i.e., PIMA INDIAN DIABETICS DATASET and CHRONIC KIDNEY DISEASE DATASET downloaded from the Kaggle repository are used. Consequently, there are two phases to this study. One is data sampling and the other one is the Prediction model. Various data sampling techniques, including SMOTE, K-means SMOTE, SMOTENC (SMOTE for Nominal and Continuous), and SMOTEN (SMOTE for Nominal), were employed in this study. To apply the various ML (Machine learning) algorithms, after obtaining updated data sampling datasets, in order to forecast the early stages of diabetes and chronic kidney disease. The optimum sampling and machine learning methods are determined based on the implemented, validated accuracy, precision, and recall values.