Causal-Aware Federated Neuro-Semantic Capsule Transformer Framework for Robust and Early Prediction of Meningitis

Meningitis is a severe neuroinfectious disease that can cause serious neurocognitive impairment and incurable rates of death when it is not diagnosed early. Existing diagnostic and forecasting techniques rely on manual clinical examination and laboratory analyses, which are timeintensive, error-prone, and, in most cases, yield high rates of false positives. To circumvent these shortcomings, this paper proposes next-generation deep learning and optimisation approaches within a higher-level artificial intelligence framework to predict meningitis with high accuracy and at anearly stage. The proposed model incorporates a NeuroSemantic Capsule Transformer Network (NSCTN), which uses a combination of a capsule-based spatial hierarchy andtransformer-based contextual learning to learn complex interactions among clinical, biochemical, and neuroimaging
features. Before classification, an Adaptive Distribution
Harmonisation (ADH) strategy is used to normalise the data,
aiming to remove scale imbalance and noise. The Deep Causal
Impact Scoring (DCIS) mechanism is used to assess feature
relevance, as it measures causal impact rather than simple
correlation. To further optimise feature selection and
convergence efficiency, an Adaptive Bayesian Hypergraph
Attention Optimiser (ABHAO) is used, which allows
modelling of multi-feature interactions while preserving
probabilistic uncertainty. Moreover, a Causal-Aware
Federated Meta Learning Network (CAFMLN) is proposed to
enhance generalisation and the robustness of heterogeneous
clinical datasets without compromising data privacy. Through
experimental confirmation, the proposed framework is shown
to be far better than traditional machine learning, deep neural networks, and state-of-the-art models in terms of accuracy, precision, recall, F1-score, and false alarm rate. The model also has better generalisation properties and robustness todifferent data distributions.