Nature-Inspired Computing and Metaheuristics Algorithms for Optimizing Data Mining Performance


Nature-inspired computing and metaheuristics (NiCam) are gaining popularity in research community for their advantages applicable in computational intelligence, data-mining and their applications. Borrowed from the wonders of nature, NiCam algorithms computationally optimize complex search problems, and they show an edge in performance and search efficiency, compared to earlier optimization techniques.

In this project, the trio-team proposed to further innovate NiCam algorithms by mainly two directions: (1) applying NiCam for solving bottlenecks in data-mining algorithms; (2) extending current NiCam algorithms to novel ones by studying the technical details in greater depth, such as mathematical analysis of algorithm structures and convergence.

The research team has competent track records in NiCam research - Yang invented Firefly algorithm in 2009, Deb has a history of innovating optimization algorithms. Recently Yang & Deb proposed Cuckoo algorithms; Fong formulated Wolf Search Algorithm. In this new synergy, where the expertises of the three researchers combine, selected NiCam algorithms are to be innovated, improved and applied into several typical data-mining applications of practical importance, including global optimum clustering, feature selection for reducing the high data dimensionality in classification model induction and optimizing Class-balancing. Automatic parameters optimization by NiCam for data-mining will be studied.

The prime objective is to innovate new NiCam algorithms which are to be infused with data mining algorithms; the secondary objective is to apply them to data mining applications. Innovation is to be achieved by four means: –

(1) Originate new algorithm(s) from inspiration of nature or biological entities’ movements; (The PI and co-PIs have track-records in achieving so. c.f.

(2) Further improve from a selection of NiCam algorithms which are originally invented by the PI and co-PIs recently.

(3) Apply new NiCam algorithms for solving data-mining algorithm bottlenecks, such as global optimization, feature selection for reducing the high dimensionality of the multivariate training dataset, finding the best set of parameter values with respect to possibly multiple performance indicators. Each of these are NP-hard problems in data mining.

(4) Automatic parameter tuning and efficiency improvement in data mining application by using and may be modifying NiCam algorithms, will be an important objective for this research. For different data mining algorithms in different applications, we need different settings of NiCam algorithms which should be automated as an integrated algorithm.

With the achievement of innovating new NiCam algorithms, especially those that could integrate with data-mining for performance improvement, the expected significance is the intellectual contribution and knowledge contribution to computer science.