Determining neural network architecture using data mining techniques
Mohamed Lafif Tej, Ştefan Holban · 2018
This paper presents techniques used to determine the optimal neural network architecture using pattern recognition and data mining. Clustering techniques highlight a number of common characteristics of input forms, which are classified into groups based on a given criterion. In the proposed method, the number of groups obtained using clustering techniques on the training data of a neural network represents the main factor for determining the optimal number of hidden layers for a multi-layer neural network. Use of this method allows the design of an optimal neural network to be unsupervised and will decrease its build time.