Abstract
This article introduces the J-score, a heuristic feature selection technique capable of selecting a useful subset of attributes from a dataset of potential inputs. The utility of the J-score is demonstrated through its application to a dataset containing historical information that may influence the house price index in the United Kingdom. After selecting a subset of features deemed appropriate by the J-score, a predictive model is trained using an artificial neural network. This model is then tested and the results compared with those from an alternative model, built using a subset of features suggested by the Gamma test, a non-linear analysis algorithm that is described. Other control subsets are also used for the assessment of the J-score model quality. The predictive accuracy of the J-score model relative to other models provides evidence that the J-score has good potential for further practical use in a variety of problems in the feature selection domain.
Original language | English |
---|---|
Pages (from-to) | 136-153 |
Number of pages | 18 |
Journal | Neural Computing and Applications |
Volume | 15 |
Issue number | 2 |
DOIs | |
Publication status | Published - 1 Apr 2006 |
Keywords
- Artificial neural network
- Attribute selection
- C63: computational techniques
- Gamma test
- J-score