Research on optimizing network intrusion detection using deep learning and big data in intelligent elderly care

Abstract

Internet of Things devices are increasingly embedded in elderly care services, expanding exposure to network intrusions that can disrupt remote monitoring and compromise sensitive data. This study develops a stacked deep-learning intrusion detection meta-model for elderly care network settings and evaluates it using the Network Security Laboratory–Knowledge Discovery and Data Mining (NSL-KDD) and Canadian Institute for Cybersecurity Intrusion Detection System 2018 (CICIDS2018) datasets. The approach integrates deep neural networks, convolutional neural networks, recurrent neural networks with long short-term memory and gated recurrent units, and autoencoders by fusing their calibrated decision outputs in a second-stage learner. Data preprocessing included encoding of categorical attributes, normalization, and class-imbalance handling, with model comparison performed using five-fold cross-validation and one-way analysis of variance with Tukey’s post hoc contrasts. The proposed meta-model achieved 99.85% accuracy, 99.2% precision, 99.1% recall, and a 99.15% F1 score, exceeding individual base learners and comparator ensembles, and showed strong detection for frequent service-disruption and reconnaissance attacks while remaining less sensitive to rare exploit categories (approximately 0.85 precision/recall for low-support classes). These results indicate that decision-level fusion can improve robustness under class imbalance and supports low-latency deployment in resource-constrained care facilities when implemented in an edge–cloud monitoring workflow.