Hybrid Classification Algorithms With Application In Quality Improvement

Quality is essential for manufacturing industry. The manufacturing industry has come toa new age of informatization, intellectualization and networking. The quality analysis andimprovement is the prediction and classification of the mass, high-dimensional and complexdata. The dissertation focuses on the research of classification algorithms of quality analysisand improvement.The main contents and contributions of the dissertation are as follows:1. ISOMAP kernel based hybrid manifold learning and support vector machinesalgorithmAn Isometric mapping (ISOMAP) kernel based hybrid manifold learning and supportvector machines algorithm (IKML-SVM) is proposed to deal with the hybrid algorithm ofdimension reduction and classification of the mass, high-dimensional and complex qualitydata. In the IKML-SVM algorithm, ISOMAP and support vector machines are connected witha kernel space. The kernel function induced form the kernel space called ISOMAP kernelfunction (IKF) is proposed. The ISOMAP kernel function meets the Mercer condition isproved. Theorems related to the ISOMAP kernel functions are proved. Simulation results withUCI data sets shown that the IKML-SVM is efficient, precise and less data quality depended.2. Dissimilarity based kernel space for support vector machines algorithmA dissimilarity based kernel space for support vector machines algorithm (DKS-SVM) isproposed for the complimentary of the IKML-SVM to deal with the situation that thedissimilarity can’t compute by Euclidean distance. An additive constant method is used inDKS-SVM to build the kernel space. The kernel function is proposed for the kernel space.The DKS-SVM is validated by using the data sets form UCI repository.3. Decision analysis algorithm based on manifold learningIKML-SVM and DKS-SVM are focused on the low-dimension embedding and they areprediction algorithms. A decision algorithm should be proposed to extract the decision rules.The decision analysis algorithm based on manifold learning (DAML) is proposed to analysisthe input data sets. The Support subset significant algorithm based on equivalence relation(S3ER) in DAML can compute the significant, discrimination and support subset of theconditional attribution value to the decisional attribution value. The validity of the DAML isverified by UCI data sets.4. A case study of an aviation enterprise A case study of quality analysis and improvement based on the real quality data from anaviation enterprise is presented. The data collection, data pretreatment, data mining andknowledge discovery are shown step by step in the case study. The IKML-SVM and theDAML is combined together with the support vectors as the quality prediction and knowledgediscovery. The support vectors extracted from the IKML-SVM are the input of the DMAL.The quality data is reduced by vertical and horizontal direction, so it can extract qualityimprovement rules easily.