In this book, the Fast Recursive Algorithm (FRA) and Two-Stage Selection (TSS) methods proposed by Prof. Li and Prof. Irwin have been improved to integrate Bayesian regularisation to prevent over-fitting and leave-one-out cross validation for automatic model construction. To further enhance model generalization capability, some heuristic methods were also embedded in the two-stage selection to optimize the non-linear parameters involved in subset model construction. These include Particle Swarm Optimization (PSO), Defferential Evolution (DE), and Extreme Learning Machine (ELM). The effectiveness and efficiency of all these advanced methods have been confirmed on both well-known benchmarks and real world data sets from automotive engine and polymer extrusion applications.