On the quality of learning of root-based decision making of partially connected neural networks under conditions of limited data

Abstract

On the quality of learning of root-based decision making of partially connected neural networks under conditions of limited data

Alekseev A.O., Kozhemyakin L.V.

Incoming article date: 30.01.2024

The quality of training of incompletely connected neural networks based on decision's roots is discussed. Using the example of limited data on patients with clinically diagnosed Alzheimer's disease and conditionally healthy patients, a decision's root and the corresponding neural network structure are found by preprocessing the data. The results of training an incompletely connected artificial neural network of this type are demonstrated for the first time. The results of training of this type of neural network allowed us to find a neural network with an acceptable level of accuracy for the practical application of the obtained neural network to support medical decision making - in the considered example for the diagnosis of Alzheimer's disease.

Keywords: neural networks, complex assessment mechanisms; decision roots, criteria trees, convolution matrices, data preprocessing