A methodology to explain neural network classification

Authors:
Raphael Féraud;Fabrice Clérot
Affiliations:
France Télécom R&D, 2, avenue Pierre Marzin, 22300 Lannion, France;France Télécom R&D, 2, avenue Pierre Marzin, 22300 Lannion, France
Venue:
Neural Networks
Year:
2002

Citing 13
Cited 12

Self-organization and associative memory: 3rd edition

Self-organization and associative memory: 3rd edition
Learning internal representations by error propagation

Parallel distributed processing: explorations in the microstructure of cognition, vol. 1
Optimal brain damage

Advances in neural information processing systems 2
On the relations between discriminant analysis and multilayer perceptrons

Neural Networks
Interpreting neural-network connection weights

AI Expert
Stock performance modeling using neural networks: a comparative study with regression models

Neural Networks
The nature of statistical learning theory

The nature of statistical learning theory
Extraction of Logical Rules from Neural Networks

Neural Processing Letters
Derivation of the multilayer perception weight constraints for direct network interpretation and knowledge discovery

Neural Networks
Bootstrapping confidence intervals for clinical input variable effects in a network trained to identify the presence of acute myocardial infarction

Neural Computation
An empirical measure of element contribution in neural networks

IEEE Transactions on Systems, Man, and Cybernetics, Part C: Applications and Reviews
New algorithms for learning and pruning oblique decision trees

IEEE Transactions on Systems, Man, and Cybernetics, Part C: Applications and Reviews
A connectionist approach to generating oblique decision trees

IEEE Transactions on Systems, Man, and Cybernetics, Part B: Cybernetics

Evaluation and classification of otoneurological data with new data analysis methods based on machine learning

Information Sciences: an International Journal
Prediction of corporate financial health by Artificial Neural Network

International Journal of Electronic Finance
The interpretation of feedforward neural networks for secondary structure prediction using sugeno fuzzy rules

International Journal of Hybrid Intelligent Systems - VIII Brazilian Symposium On Neural Networks
Selective enhancement learning in competitive learning

IJCNN'09 Proceedings of the 2009 international joint conference on Neural Networks
Building credit scoring models using genetic programming

Expert Systems with Applications: An International Journal
Information-theoretic enhancement learning and its application to visualization of self-organizing maps

Neurocomputing
How to Explain Individual Classification Decisions

The Journal of Machine Learning Research
Modelling complex data by learning which variable to construct

DaWaK'10 Proceedings of the 12th international conference on Data warehousing and knowledge discovery
Selective information enhancement learning for creating interpretable representations in competitive learning

Neural Networks
A hybrid approach to integrate genetic algorithm into dual scoring model in enhancing the performance of credit scoring model

Expert Systems with Applications: An International Journal
Decision tree-based technology credit scoring for start-up firms: Korean case

Expert Systems with Applications: An International Journal
Driven forward features selection: a comparative study on neural networks

ICONIP'06 Proceedings of the 13th international conference on Neural Information Processing - Volume Part II

Quantified Score

Hi-index	0.01

Visualization

Abstract

Neural networks are still frustrating tools in the data mining arsenal. They exhibit excellent modelling performance, but do not give a clue about the structure of their models. We propose a methodology to explain the classification obtained by a multilayer perceptron. We introduce the concept of 'causal importance' and define a saliency measurement allowing the selection of relevant variables. Once the model is trained with the relevant variables only, we define a clustering of the data built from the hidden layer representation. Combining the saliency and the causal importance on a cluster by cluster basis allows an interpretation of the neural network classifier to be built. We illustrate the performances of this methodology on three benchmark datasets.