作者: Morten Goodwin , Torry Tufteland , Guro Ødesneltvedt , Anis Yazidi
DOI: 10.1007/S11721-017-0145-6
关键词:
摘要: Ant colony optimisation (ACO) for classification has mostly been limited to rule-based approaches where artificial ants walk on datasets in order extract rules from the trends data, and hybrid which attempt boost performance of existing classifiers through guided feature reductions or parameter optimisations. A recent notable example that is distinct mainstream PolyACO, a proof-of-concept polygon-based classifier resorts ACO as technique create multi-edged polygons class separators. Despite possessing some promise, PolyACO significant limitations, most notably, fact supporting only two classes, including features per class. This paper introduces PolyACO+, an extension three ways: (1) PolyACO+ supports classifying multiple (2) dimensions enabling with more than features, (3) substantially reduces training time compared by using concept multi-levelling. empirically demonstrates these updates improve algorithm such degree it becomes comparable state-of-the-art techniques SVM, neural networks, AntMiner+.