Comparative Analysis of Machine Learning Algorithm for Classification of different Osteosarcoma types

Abstract

Histopathological osteosarcoma data analysis gives a lucrative way to study the pathological texture of Osteosarcoma. Osteosarcoma is classifying into Viable, Necrotic, and Non-Tumor. A Viable cell indicates a cancerous cell element and a necrotic cell means the cancerous cell nuclei kill by chemotherapy or radiation therapy. Necrosis is the process where cell injury results in the death of a cell. Classification of Osteosarcoma is a very time-consuming and complicated task due to its inter-class similarities and variations. Improvement in Machine Learning(ML) algorithm and Graphical Processing Unit(GPU) gives more accurate results for the classification and prediction of Osteosarcoma. In this paper, we use four ML algorithms for the classification of OST: Decision Tree(DT), Support Vector Machine(SVM), K-Nearest Neighbors(KNN) and AdaBoost(Adaptive Boosting). The ML model performance evaluation uses performance metrics like Accuracy, Sensitivity, specificity, F1 -score, Kappa, and AUC (area under the curve). All four classifiers successfully classified Osteosarcoma into three types. The overall accuracy of DT, SVM, KNN and Adboost is 81.22%, 83.80%, 86.90% and 91.70% respectively. Adaboost algorithm outperforms the other three algorithms with overall accuracy 91.70 %, sensitivity 91.60 %, specificity 96.00%, F1 score 90.60%, Kappa 0.87 and AUC 0.99.