A Multiview Discriminant Feature Fusion-Based Nonlinear Process Assessment and Diagnosis: Application to Medical Diagnosis

Fusion of large-scale information is the key strategy for the complete understanding of many nonlinear and complicated industrial and medical processes. This paper presents the proposed multiview (MV) feature fusion-based learning generalizing discriminant correlation analysis (DCA) for assessment and diagnosis of nonlinear processes. The core focus of this algorithm is to explore the effectiveness of MV information embedding in learning models and viable implementation in real-time applications. Our method is capable of incorporating high-dimensional information inherent in MV features generated from available inputs using the proposed DCA-based scheme. The algorithm is tested with two real-time electromyogram data sets, which include three categories of nonlinear data-amyotrophic lateral sclerosis, myopathy, and healthy control subjects. A set of MV features is generated in both the time and the wavelet domain for all of the study groups. The features are subjected to DCA projection and optimization and obtained low-order features are concatenated using DCA-based fusion scheme. The discriminant features are applied for the statistical analysis and the model validation. The model achieves an accuracy of 99.03% with a specificity of 99.58% and sensitivities of 98.50% and 97.59%. However, the accuracy over the second data set is 100% with sensitivities of 100% and the specificity of 100%. Results are further compared with the state-of-the-art methods. The proposed scheme is promising and outperforms many state-of-the-art methods, and thus it ensures the faithfulness for industrial applications.