Over-Fitting in Ensembles of Neural Network Classifiers Within ECOC Frameworks

Prior, Matthew; Windeatt, Terry

doi:10.1007/11494683_29

Cited by 10 publications

(9 citation statements)

References 9 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Even though different aspects have been addressed in studies investigating overfitting in the context of classifier ensembles, for instance regularization terms [24] and methods for tuning classifier members [20], very few authors have focused on proposing methods to tackle overfitting at the selection stage. Tsymbal et al [34] suggested that using individual member accuracy (instead of ensemble accuracy) together with diversity in a genetic search can overcome overfitting.…”

Section: Introductionmentioning

confidence: 99%

Overfitting cautious selection of classifier ensembles with genetic algorithms

2009

View full text Add to dashboard Cite

Section: Introductionmentioning

confidence: 99%

Overfitting cautious selection of classifier ensembles with genetic algorithms

2009

View full text Add to dashboard Cite

“…However, aside from manipulating the outputs of the binary classifiers which is the key idea of ECOC, the diversity has been seldom referred in the ECOC study yet. To our knowledge, only in [10], Prior and Windeatt manipulated different parameter settings of the base classifiers (multi-layer perceptrons).…”

Section: The Proposed Ecocmentioning

confidence: 99%

Weight optimization and layered clustering-based ECOC

Zhang

2013

2013 IEEE International Conference on Acoustics, Speech and Signal Processing

View full text Add to dashboard Cite

Error correcting output code (ECOC) is a general framework of solving a multiclass classification problem via a binaryclass classifier ensemble. In this paper, we propose a new heuristic coding method, named weight optimization and layered clustering-based ECOC (WOLC-ECOC). It iterates the following two steps until the training risk converges. The first step employs the layered clustering-based approach [1]. The approach can construct multiple different strong binaryclass classifiers on a given binary-class problem, so that the heuristic training process will not be blocked by some difficult binary-class problems. The second step is the weight optimization technique [2]. It guarantees the non-increasing of the heuristic training process whenever we add new classifiers to the ECOC ensemble. Experimental results on several benchmark sets demonstrate that WOLC-ECOC is more effective than 15 referenced coding-decoding ECOC pairs.

show abstract

“…However, unlike channel codes in communication, the "channels" in ECOC are influenced by the bipartitions of classes: if the classes are partitioned improperly, the "noise" (errors) of the channels may be rather high. Furthermore, because there are only 2 P−1 −1 possible bipartitions in any binary codes, the code length is limited when P is small [34]. Finally, the errorcorrecting ability of ECOC is severely limited.…”

Section: A Survey On the Coding Phasementioning

confidence: 99%

“…Hatami [55] tried to delete the columns of a coding matrix that have weak diversities. 2) Other types of diversity were seldom explored: Only Prior and Windeatt [34] manipulated different parameter settings of multilayer perceptrons. Bagheri et al [39], [40] trained different base dichotomizers with different feature subsets.…”

Section: A Survey On the Coding Phasementioning

confidence: 99%

“…The second class is problem-dependent codings [19]- [37] which aim to solve given multiclass problems without considering the errorcorrecting ability of coding matrices much. This class attracted much attention in recent years, such as discriminant ECOC (DECOC) [26] ECOC-optimizing node embedding (ECOC-ONE) [27], [38] subclass-ECOC [28] manipulations of features [39], [40] and manipulations of the parameters of base dichotomizers [34]. The decoding methods are various distance metrics, including hamming distance (HD), Euclidean distance (ED), probabilistic [41] loss-based (LB) [42] and loss-weighted (LW) [43] decodings.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Heuristic Ternary Error-Correcting Output Codes Via Weight Optimization and Layered Clustering-Based Approach

Zhang

2015

IEEE Trans. Cybern.

View full text Add to dashboard Cite

One important classifier ensemble for multiclass classification problems is error-correcting output codes (ECOCs). It bridges multiclass problems and binary-class classifiers by decomposing multiclass problems to a serial binary-class problems. In this paper, we present a heuristic ternary code, named weight optimization and layered clustering-based ECOC (WOLC-ECOC). It starts with an arbitrary valid ECOC and iterates the following two steps until the training risk converges. The first step, named layered clustering-based ECOC (LC-ECOC), constructs multiple strong classifiers on the most confusing binary-class problem. The second step adds the new classifiers to ECOC by a novel optimized weighted (OW) decoding algorithm, where the optimization problem of the decoding is solved by the cutting plane algorithm. Technically, LC-ECOC makes the heuristic training process not blocked by some difficult binary-class problem. OW decoding guarantees the nonincrease of the training risk for ensuring a small code length. Results on 14 UCI datasets and a music genre classification problem demonstrate the effectiveness of WOLC-ECOC.

show abstract

Over-Fitting in Ensembles of Neural Network Classifiers Within ECOC Frameworks

Cited by 10 publications

References 9 publications

Overfitting cautious selection of classifier ensembles with genetic algorithms

Overfitting cautious selection of classifier ensembles with genetic algorithms

Weight optimization and layered clustering-based ECOC

Heuristic Ternary Error-Correcting Output Codes Via Weight Optimization and Layered Clustering-Based Approach

Contact Info

Product

Resources

About