Fused activation deep neural network classifier with Cat-Hawk optimization for fetal health classification

Accurate and timely evaluation of fetal health is essential for appropriate clinical intervention and better neonatal outcomes. Traditional CTG reading is usually prone to subjectivity and variability. Conventional machine learning techniques tend to fail while addressing the nonlinear characteristics of cardiotocography (CTG) data, particularly under conditions of class imbalance and sparse features. To overcome these limitations, in this work, a new Fused Activation Deep Neural Network (FADNN) optimized by the Cat-Hawk Optimization (CHO) algorithm is introduced for the fetal health classification task. The architecture of the proposed FADNN model merges several activation functions like ReLU, LeakyReLU, and Swish into parallel branches, encouraging rich feature learning and enhancing model generalization. Feature selection and hyperparameter tuning are performed using the CHO algorithm, which enhances model performance effectively. The proposed CHO-based FADNN model with the SMOOTENN data balancing technique achieved a remarkable accuracy, precision, recall, and specificity of almost 99% and an AUC of 100%. Ablation study further verifies the dominance of the CHO-based optimization approach compared to other metaheuristics such as PSO, CSO, and HHO. The developed framework provides a practical and interpretable solution to intelligent fetal monitoring with strong potential to support clinicians in real-time fetal health assessment.