DETERMINING OPTIMAL NEURAL NETWORK PARAMETERS FOR PREDICTING THE SUCCESS OF IT PROJECTS

Abstract

The growing importance of IT projects in the modern economy necessitates the application of advanced analytical tools for assessing their potential success. This study is devoted to determining the optimal parameters of neural network models for predicting the success of IT projects. The main objective is to improve the quality, reliability, and stability of classification results through the justified selection of neural network architecture and its hyperparameters.

The research is based on a dataset of Ukrainian IT projects, which ensures the relative stability of the economic environment and minimizes the impact of structural shifts. A set of key factors influencing project success was identified, including financial indicators, user activity metrics, and digital presence characteristics. The success of a project is represented as a binary variable, allowing the problem to be formalized as a binary classification task.

Multilayer perceptron models were used as the primary modeling tool, while radial basis function networks were also tested but demonstrated insufficient performance. A series of computational experiments was conducted to determine the optimal number of neurons in the hidden layer. The results indicate that although models with fewer neurons provide higher average accuracy, models with a larger number of neurons ensure greater stability. Considering both accuracy and variability measures, the optimal configuration was identified as a neural network with 12 hidden neurons.

Furthermore, the study investigates the potential of ensemble approaches, which combine several neural network models with different configurations. The findings confirm that ensemble structures improve both predictive accuracy and robustness of classification results.

The obtained results contribute to the development of methodological approaches for optimizing neural network parameters in IT project success prediction and can be applied in decision-making processes by investors, venture funds, and project managers.