The analysis of physiological data often requires effective feature extraction and classification techniques from complex and high-dimensional physiological data. Existing approaches like support vector machine, K-nearest neighbor, and random forest suffer from issues such as overfitting, limited generalization, and computational inefficiency while dealing with large datasets. Moreover, traditional methods might fail to preserve important variance from high-dimensional data, which results in suboptimal anomaly detection and classification. It does point out the requirements of improved methods that would integrate some dimensional reduction, robust cross-validation, and powerful deep models, which could enhance its overall performance. The principal component analysis with K-fold and deep neural network (DNN) method is proposed to reduce problems by incorporating PCA feature extraction, reducing the dimension complexity while retaining the major part of the variance it. K-fold cross-validation is used in such a way that both proper training and testing minimize the overfitting to achieve better generalization ability. By using DNNs, the framework obtains advanced patterns in the input and achieves excellent results such that it attains 99.1% accuracy with a recall of 98.5% and F-score equal to 98.1%. The results therefore tend to outperform typical models, making the devised framework a highly efficient and reliable mechanism toward anomaly detection and classification with physiological datasets. © 2025 Elsevier B.V., All rights reserved.