Adaptive Synthetic-Nominal (ADASYN-N) and Adaptive Synthetic-KNN (ADASYN-KNN) for Multiclass Imbalance Learning on Laboratory Test Data

Abstract

Annually about 1,500 cases of cervical cancer are found in Indonesia, which made Indonesia as the country with the highest number of cervical cancer cases in the world. Cervical cancer screening and HPV testing are done with a Pap smear test. However, this examination requires a lot of time, costly and highly susceptible bias of the observer during the process of investigation and analysis. To overcome these problems, several studies have modeled the machine learning with a variety of approaches have been made. However, these studies are constrained by the limitation of the data amounts and the imbalanced data that caused by the different ratio of each case. This can lead to errors in the classification of the minority due to the tendency of the classification results that focus on the majority class. This study addressed the handling imbalance data on classification of cases Pap test results using the method of over-sampling. ADASYN-N and ADASYN-KNN algorithms were proposed as a development of ADASYN algorithm to handle datasets with nominal data types. This study included SMOTE-N algorithm to deal with the problem as comparison algorithm. As the results, ADASYN-KNN with the preference “0” gave the highest accuracy, precision, recall and f-score of 95.38%; 95.583%; 95.383%; and 95.283%. The highest ROC area value was obtained with the ADASYN-KNN with preference “1” of 99.183%.