An Experimental Study of Walrus Optimization-Based Deep Learning Models for Software Effort Estimation

Abstract

Software Effort Estimation plays a pivotal role in project planning, resource allocation, and cost management within the software development life cycle. This study investigates the integration of deep learning models Convolutional Neural Network, Recurrent Neural Network, Long Short-Term Memory, Bidirectional Long Short-Term Memory and Multilayer Perceptron with the Walrus Optimization Algorithm for hyperparameter tuning. By using five standard SEE datasets including China, Kemerer, Kitchenham, Maxwell, Miyazaki94, we assess the model's performance using multiple evaluation metrics: Mean Squared Error, Mean Absolute Error, Coefficient of Determination (R2), Mean Magnitude of Relative Error, and Prediction Accuracy at 2 5 % (Pred(25)). Experimental results reveal that LSTM and its variant consistently outperform other models in terms of MSE, MAE, and R2 in most datasets, demonstrating the advantage of sequence-based learning for SEE. The proposed framework offers a replicable methodology for applying evolutionary optimization to deep learning in SEE.