Learning Medical Diagnosis via Scaled Convex Hull-Based SK Algorithm
Yuqing Liu, Qiangkui Leng, Shurui Wang · 2019 IEEE 8th Data Driven Control and Learning Systems Conference (DDCLS) · 2019
Machine learning algorithms, especially support vector machines (SVMs), were from the very beginning designed and used to analyze medical datasets. As a geometric dual representation of SVMs, Schlesinger-Kozinec (SK) algorithm achieves classification by solving the nearest point pair between convex hulls. However, the initial SK algorithm is only for linear separable problems. For non-separable problems, it needs to perform appropriate convex hull transformation, i.e., so-called scaled convex hull (SCH). This paper first gives the theory foundation of scaled convex hull. Then, the SK algorithm as well as its kernelization process is further explained. At last, we employ the SCH and the kernel SK (KSK) algorithm to carry out medical diagnostic tasks. The results show that compared with other well-known machine learning algorithms, the SCH-based KSK algorithm can achieve higher prediction accuracy.