Robustness of data mining tools under varying levels of noise: Case study in predicting a chaotic process

Many processes in the industrial realm exhibit stochastic and nonlinear behavior. Consequently, an intelligent system must be able to adapt to nonlinear production processes as well as probabilistic phenomena. In order for a knowledge based system to control a manufacturing process, an important capability is that of prediction: forecasting the future trajectory of a process as well as the consequences of the control action. This paper examines the robustness of data mining tools under varying levels of noise while predicting nonlinear processes, including chaotic behavior. The models evaluated include the perceptron neural network using backpropagation, the recurrent neural network and case based reasoning. The concepts are crystallized through a case study in predicting a chaotic process in the presence of various patterns of noise.