Analytical Decision Boundary Feature Extraction for Neural Networks with Multiple Hidden Layers

Go, Jinwook; Lee, Chulhee

doi:10.1007/978-3-540-45179-2_71

Cited by 2 publications

(2 citation statements)

References 4 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The approach is said to be Bayesian, since the classifier is designed according to an algorithm for the minimization of the average misclassification risk, which allows getting a locally optimal approximation of the Bayes decision border, from which the most relevant features can be derived. In the same family, we have: Analytical Decision Boundary Feature Extraction (ADBFE) [9], which is based on a multilayer perceptron and SVM Decision Boundary Analysis (SVMDBA) [22], exploiting support vector machines. EDBFE-based family of algorithms requires, however, expertise for setting algorithm parameters and more computational time than LDA-based ones.…”

Section: B Existing Solutions For Feature Rankingmentioning

confidence: 99%

“…Another family of FE algorithms for classification is the one based on the EDBFE method [3,9,22], where classification algorithms developed in the data mining field are exploited for defining the decision border; then, directions normal to the decision border are analytically extracted and used to build the EDBFM matrix. The main characteristic of …”

Section: B Existing Solutions For Feature Rankingmentioning

confidence: 99%

See 1 more Smart Citation

A novel feature ranking modelling in GIS context: Addressing complexity and cost issues

Gemelli

Diamantini

Potena

2009

2009 17th International Conference on Geoinformatics

View full text Add to dashboard Cite

nowadays, there is the trend to carry out decisions and analysis on geospatial data by a massive computational approach. The amount of geospatial information available is increasing exponentially as result of the increasing interoperability between informative systems. In a multiplicity of applications and services spatial decision is carried out to pursue business goals, often without involving experts in geography. The informative systems have an increasing autonomous decisional capability on information selection and analysis. The demand is to have systems that require only an input goal, and produces decisions that humans can understand and integrate with their own decisions. In this paper it is proposed an automatic method of feature ranking, which can sort a heterogeneous set of features by their importance in accomplishing an analytical goal. This method produces a rank model that helps to select the minimal set of features needed to pursue a goal with a wanted accuracy or resources involvement. This feature ranking is expected to supports fundamental decisional making in elaborating geospatial data. The method is based on data mining algorithms; the obtained rank model appears to be spatially scalable and fits well to human form of knowledge.

show abstract

Section: B Existing Solutions For Feature Rankingmentioning

confidence: 99%

Section: B Existing Solutions For Feature Rankingmentioning

confidence: 99%