Soft Tissue Tumor Classification using Stochastic Support Vector Machine

Abstract

As healthcare is becoming one of the most rapidly changing industries by the increasing type of diseases, technology plays an important role in helping medical staffs solve those medical problems. Soft tissue tumors are tumors in the musculoskeletal system that involve soft tissue (tissue other than bone tissue). It includes muscle tissue, nerves, blood vessels, fat, and connective tissue. This soft tissue tumor is divided into two, namely benign and malignant. To prevent any medical errors in classifying patients’ data, one machine learning called Stochastic Support Vector Machine is being studied. In this study, we will evaluate soft tissue tumor patients’ data in Nur Hidayah Hospital, Yogyakarta, Indonesia using Stochastic Support Vector Machine to see its accuracy. The result is encouraging that Stochastic Support Vector Machine works better than the original Support Vector Machine.

Improvement in automated diagnosis of soft tissues tumors using machine learning

“…The comparison based on the nonparametric statistical test of the performance of DT and SVM algorithms with other classification algorithms showed that DT or SVM algorithms worked just as well or even better. Compared to the previous work done by the stochastic-SVM algorithm approach on this same dataset [34] , we have a significant performance improvement of more than 32.9% and 26.3% in terms of accuracy and f1-measure, respectively. Moreover, we corrected several anomalies related to the process of the construction of classification models, which were recorded in this paper.…”

“…Most recently, Rustam et al [10] published additional work on this problem by improving the performance of Ref. [34] with Fuzzy C-means clustering and an SVM model, but the performance was still weaker compared to ours. Overall, we have improved the best current values (71.43% for accuracy and 83.33% for f1-measure) [10] on this database by 27.57% and 15.67 % in terms of accuracy and f1-measure, respectively.…”

“…The size of this dataset always represents a small random sample of the entire population of STT. Just like other researches [10,34] , we forced to resample these data in order to build the best model, but we detailed this resampling process supported by cross-validation of the results. We needed much larger training samples for practical applications to cover all the different pathological types of STT [2] .…”

“…In Table 4, we note that compared to the previous [34] 57.82 -66.00 Stochastic-SVM [34] 64.80 -72.00 Fuzzy C-means [10] 71.43 -81.82 SVM [10] 71.43 -83.33 study of Zahras et al [34] , which also used the SVM and stochastic-SVM algorithms, there is massive improvement in the performance. In terms of accuracy and f1-measure, our model yielded an increase of 32.9% and 26.3%, respectively, above their SVM model.…”

“…Since the objective of an automatic classification model is to predict results as close to reality as possible, the main challenge of our study, we compared our results to the previous work carried out by Zahras et al [34] on this same problem and summarized this comparison in Table 4.…”