Study of anomaly registration detection based on multilayer kernel autoencoder extreme learning machine model

The deep integration of artificial intelligence (AI) and health informatics has driven the digital transformation of hospital services, but improper registration practices (scalpers) exacerbate the uneven allocation of appointment resources. The paper proposes an integrated machine learning algorithm DKELM-PSS for advanced feature mining of high-dimensional massive health information system (HIS) data in hospital to enable detection of anomalous registration behavior. First, Sparse Principal Component Analysis (Sparse PCA) is employed to preprocess the raw data, thereby performing data denoising and dimensionality reduction. Then, the deep kernel extreme learning machine model (DKELM) is constructed for deep feature extraction by stacking multiple kernel self-coders (KELM-AE) for multilayer forward coding. Kernel-based mappings in the hidden space enhance the linear separability of the raw data in high-dimensional feature space. Unsupervised extraction of deep features by autoencoder structure for advanced feature representation learning. The optimal parameter configuration of DKELM is realized by Salp Swarm Algorithm (SSA) to improve the classification accuracy and model stability. Compared with traditional detection methods such as SVM-RBF, XGBoost, and ResNet, experimental results validate that DKELM-PSS attains the optimal accuracy of 0.9942 on HIS anomaly detection, corroborating its effectiveness and robustness. The paper proposes an efficient anomaly detection method, which is conducive to optimizing the allocation of medical resources and promoting the intelligent development of hospitals.