Experimental Evaluation of Lupus Disease Diagnosis using Hybrid Neural Learning and Classification Scheme

Systemic Lupus Erythematosus (SLE) causes difficulties in early diagnosis because of its multi-systemic impacts and inconsistent recognition patterns among patients. The proposed research incorporates a hybrid neural learning and classification system which unites Convolutional Neural Networks (CNNs) with Bidirectional Long Short-Term Memory (Bi-LSTM) alongside an attention mechanism to deliver early and precise Lupus diagnosis. A validated medical dataset supplies the model with clinical and laboratory features that include ANA titers and anti-dsDNA levels and ESR and CRP and complement proteins (C3, C4). The preprocessing stage involves three steps which include KNN imputation and Z-score normalization and SMOTE-based class balancing. The method uses Mutual Information combined with Recursive Feature Elimination (RFE) to select features that reduce dimensions through Principal Component Analysis (PCA). A stratified 10-fold cross-validation method evaluates the model after Bayesian hyperparameter optimization which used the Adam optimizer for training purposes. The proposed architecture reaches 97.81% classification accuracy together with precision at 96.92%, recall at 98.13% and AUC-ROC at 0.99 and F1-score at 97.52%. Independent testing shows that the proposed hybrid framework outperforms regular ML models together with independent deep models. The developed model shows robust diagnostic potential to support clinical early Lupus diagnosis and decision-making processes.