Comparative Analysis of Neural Network Architectures for Software Cost Estimation

Abstract

Software cost estimation is a critical aspect of project management, with significant implications for decision-making and resource allocation. In recent years, the use of neural networks (NNs) has shown promise in improving the accuracy of cost estimation models. This research, unique in its systematic investigation of various NN architectures, aims to determine their strengths and weaknesses in the context of cost estimation. The study focuses on three primary neural network architectures: feedforward neural networks (FNNs), recurrent neural networks (RNNs), and convolutional neural networks (CNNs). It scrutinizes each architecture's ability to capture and model the complex relationships within software development datasets, with a particular emphasis on their performance in terms of prediction accuracy, scalability, and adaptability to diverse project scenarios. By providing a comprehensive understanding of the nuanced dynamics between neural network structures and their effectiveness in software cost estimation, this review aims to offer practical insights that can guide practitioners and researchers in selecting the most suitable neural network architecture for specific cost estimation challenges. Additionally, it identifies potential areas for future research, thereby offering a roadmap for advancing the application of neural networks in software cost estimation.