Оптимізація алгоритмів сортування за допомогою нейромереж

Abstract

This paper addresses the problem of optimizing the QuickSort algorithm using artificial neural networks. Since QuickSort's performance heavily depends on the choice of the pivot element, a novel approach is proposed in which a neural network predicts the optimal pivot based on array characteristics. A multilayer perceptron (MLP) is trained on a dataset of arrays with various structures to achieve this. An experimental comparison between the standard QuickSort and its modified version with a neural network is conducted. The analysis of the results shows that using a neural network can improve the algorithm’s stability, particularly for arrays with complex structures. However, the method has limitations due to additional computational costs for training and prediction. The study results highlight the potential of neural networks in sorting tasks and open avenues for further algorithm optimization, particularly through adaptive management of recursive calls.