An enhanced neural network model with hybrid sunflower and Grey Wolf Optimization for outlier detection

Outlier detection plays a critical role in ensuring data quality and reliability across domains such as cybersecurity, healthcare, and financial analytics. Traditional neural network models often struggle with high-dimensional data, noise sensitivity, and suboptimal parameter tuning, leading to reduced detection accuracy. To address these challenges, this paper proposes an enhanced neural network model optimized using a hybrid metaheuristic approach that combines Sunflower Optimization (SFO) and Grey Wolf Optimization (GWO). The proposed Hybrid SFO-GWO algorithm leverages the exploration capability of sunflower pollination behavior and the exploitation strength of grey wolf social hierarchy to achieve optimal weight initialization and hyperparameter tuning of the neural network. The integrated model improves convergence speed, avoids local minima, and enhances detection robustness. The framework includes data preprocessing, feature normalization, hybrid optimization-based training, and anomaly scoring mechanisms. Experimental evaluation on benchmark IoT datasets demonstrates that the proposed model significantly outperforms conventional neural networks and single-optimization approaches, achieving performance values of approximately 97.3, 96.1, 95.6, and 95.9 across multiple evaluation measures. Additionally, the model exhibits faster convergence, reduced training loss, improved stability, and strong generalization capability on unseen data. These results confirm that the hybrid SFO–GWO optimized neural network provides an efficient, robust, and scalable solution for accurate outlier detection in complex data environments.