Optimizing SVM Hyperparameters using Predatory Swarms Algorithms for Use Case Points Estimation

The productivity factor is one of the key cost drivers in use case points estimation. However, this factor is uncertain because it is obtained from experts' guesses and does not depend on reliable historical data. This study aims to optimize the hyperparameters of the support vector machine technique through the use of three predatory-inspired algorithms, namely, Komodo mlipir algorithm, Grey wolf optimization, and Reptile search algorithm in predicting productivity factor rate. The optimizers searched for the best hyperparameter value to get an optimum hyperplane. In this work, we developed a hybrid model that consists of three primary stages: cluster generation using bisecting k-Medoids, prediction using an optimized support vector machine, and estimation using multiple linear regression. The hybrid model was evaluated using industrial and student projects datasets. Detailed investigations demonstrated that the komodo mlipir algorithm yielded the best mean value of 365.3 and 739.88, respectively, which has a significantly better hyperparameter of support vector machine model in predicting productivity factors among the reptile search algorithm for datasets 4 and 5 as of 0.03 (p < 0.05), but there is no significant difference among the grey wolf optimizer (p > 0.05). Meanwhile, the grey wolf optimizer obtained the best mean value of 366.47, with a significantly better hyperparameter compared with the reptile search algorithm for dataset four as of 0.047 (p < 0.05).