A Privacy Based Deep Learning Algorithm for Big Data Analytics

This thesis addresses critical challenges in privacy-preserving feature selection and classification for big data analytics. Specifically, four novel methodologies are proposed: Hierarchical Classification Feature Selection (HCFS), Privacy-Preserving Classification Selection with p-stability (PPCS), Local N-ternary Pattern combined with Modified Deep Belief Network (LNTP-MDBN), and Privacy-Preserving Cosine Similarity integrated with Multi-Manifold Deep Metric Learning (PPCS-MMDML). These approaches collectively enhance classification accuracy, optimize feature extraction from heterogeneous image sets, and robustly preserve privacy, demonstrating significant improvements in data-driven analytical applications.