A Minimal Coverage-based Classification method and its application in predictive toxicology data mining
Gongde Guo, Yu Huang · Conference proceedings/Conference proceedings - IEEE International Conference on Systems, Man, and Cybernetics · 2008
A robust method, MCC (minimal coverage-based classification), for toxicity prediction of chemical compounds is proposed. The MCC method mainly considers the local distribution of each class around a new tuple to be classified and uses minimal coverage principle - covering minimal number of tuples with different classes - to classify this new tuple. The merits of MCC over other machine learning algorithms are threefold: (1) uniform approach for both numerical and categorical data; (2) deals with missing values; (3) given a new data tuple, it provides values for all classes which measure the likelihood of the tuple being in each class. The experimental results of MCC conducted on seven toxicity data sets from real-world applications are compared with the results of IBL, DT, Ripper, MLP and SVM in terms of classification performance. This application shows that MCC is a promising method for the toxicity prediction of chemical compounds.