Extracting Classification Rules from Artificial Neural Network Trained with Discretized Inputs
- PDF / 1,499,128 Bytes
- 23 Pages / 439.37 x 666.142 pts Page_size
- 93 Downloads / 230 Views
Extracting Classification Rules from Artificial Neural Network Trained with Discretized Inputs Dounia Yedjour1 Accepted: 21 September 2020 © Springer Science+Business Media, LLC, part of Springer Nature 2020
Abstract Rule extraction from artificial neural networks remains important task in complex diseases such as diabetes and breast cancer where the rules should be accurate and comprehensible. The quality of rules is improved by the improvement of the network classification accuracy which is done by the discretization of input attributes. In this paper, we developed a rule extraction algorithm based on multiobjective genetic algorithms and association rules mining to extract highly accurate and comprehensible classification rules from ANN’s that have been trained using the discretization of the continuous attributes. The data pre-processing provides very good improvement of the ANN accuracy and consequently leads to improve the performance of the classification rules in terms of fidelity and coverage. The results show that our algorithm is very suitable for medical decision making, so an excellent average accuracy of 94.73 has been achieved for the Pima dataset and 99.36 for the breast cancer dataset. Keywords Neural network · Rule extraction · Multiobjective genetic algorithm · Discretized attributes
1 Introduction Machine learning is one of the sub-domains of artificial intelligence, aiming to extract and automatically exploit the information present in a dataset. It is used in many applications such as: medical diagnostic [1], fraud detection [2], speech recognition [3], robotics [4], etc. Artificial neural networks (ANN’s) are a part of machine learning techniques. They are often used to solve classification problems because of their ability to learn and their ability to generalize knowledge present in the training set. Several studies have been proposed to improve their performance by using the data pre-processing [5], or by training the ANN with augmented discretized variables [6] or by pruning of the units and the connections in order to have a less complex network structure [6, 7].
B 1
Dounia Yedjour [email protected] Department of Computer Science, Faculty of Mathematics and Computer Science, Université des Sciences et de la Technologie d’Oran Mohamed Boudiaf, USTO-MB, BP 1505, El M’naouer, 31000 Oran, Algeria
123
D. Yedjour
ANN’s store their knowledge in their connections weights which make their interpretation difficult by an expert. However the extraction of rules from ANN’s remains important task. Several studies have been proposed to extract rules from the ANN. The aim is to find a set of rules which mimic knowledge contained in the ANN. Local methods extract rules by checking which combination of weights can make each neuron (hidden/output) active. After extracting all rules (between the hidden layer and the output layer and between the output layer and the hidden layer), a rewriting procedure takes place which consists in generating the final rules (between the output layer and th
Data Loading...
