Several new computational intelligence algorithms and their applications are investigated in this thesis. First, a linear support vector machine decision tree (LSVMDT) is developed by building a binary tree with a linear support vector machine in each tree node. It has built-in rare event detection mechanism, and allows efficient rule extraction. Secondly, an efficient recursive update algorithm when new data becomes available is derived for least squares support vector machines. This is very essential in online learning. Thirdly, a three-layered learning system is proposed. It consists of a random mapping stage and a learning stage with ordinary least squares, error-correcting least squares, or a linear support vector machine. Next, the three-layered system with ordinary least squares is further developed into a statistical, self-organizing learning system (SSOLS). It incorporates automatic determination of the enhancement nodes using validation, VC dimension, and efficient leave-one-out methods. The t-test pruning procedure and the gradient descent update are used to make the network more compact. The last past of this thesis investigates a real-world example of how computational intelligence algorithms can be applied to automate the decision making processes in manufacturing industries. Data are collected from a Six Sigma simulator that simulates an advanced TV production line. Several computational intelligence algorithms are then used to model this manufacturing process, and a global optimization technique is applied to obtain the optimum input settings that result in maximum overall % yield. Comparison with traditional methods such as Design of Experiments shows promise in deploying computational intelligence algorithms in manufacturing enterprises of the future.
Date of this Version