Lists
Lists
Ahamed, B. Shamreen, Virdee, Bal Singh, Khanna, Ashish, Vidhya, R. G. and Sivashankar, S. (2026) Noise-resistant deep ensemble learning with optimization-driven feature fusion for heart disease diagnosis. In: 2025 International Conference on NexGen Networks and Cybernetics (IC2NC), 1-3 December 2025, Erode, India.
Heart disease is still one of the major causes of mortality across the globe, and its forecast in patients with diabetes is an onerous job owing to the intricate integration of clinical, lifestyle, and multi-source health data. Conventional Machine Learning (ML) and shallow Deep Learning (DL) models become ineffective while dealing with noisy, imbalanced, and heterogeneous datasets, which compromises their accuracy and generalizability. To overcome these difficulties, this research proposes a new Pufferfish Optimized Multi-Sensor Information Fusion Deep Ensemble Learning Network (PO-MIFDELN) for the strength and precise prediction of heart disease. The MIFDELN model integrates multi-source data by weighted fusion, convolution-pooling and attention modules for richer feature learning, and a temporal learning module to learn spatial-temporal dependencies. Ensemble stacking is utilized to merge multiple base learners, and an SVM is utilized as the meta-learner. The hyperparameters are tuned by utilizing the Pufferfish Optimization Algorithm (POA) for the optimal model performance. Experimental outcomes validate the model’s enhanced performance with accuracy of 0.980, precision of 0.970, recall of 0.990, and F1-score of 0.950 on Dataset 1; and accuracy of 0.990, precision of 0.980, recall of 0.970, and F1score of 0.980 on the Cleveland dataset. On the same values of hyperparameters (learning rate =0.001, batch size =32, dropout rate =0.2), the model showed good convergencetraining in 120 seconds and testing in 8 seconds for Dataset 1, and 65 seconds and 5 seconds respectively for the Cleveland dataset. In summary, PO-MIFDELN offers a noise-tolerant, computationally low-cost, and very reliable system for heart disease prediction among diabetics, greatly improving diagnostic reliability and accuracy over traditional approaches.
 |
View Item |