Data mining based threshold development for novelty detection

The objective of this dissertation is to develop thresholds for novelty detection with applications to statistical process control (SPC). SPC is a widely used technique for improving process and product quality. The primary tool of SPC is a control chart that is used to monitor and detect abnormal processes. Traditional control chart techniques usually require a specific distributional assumption, typically the normal distribution, to establish their control limits. However, in modern manufacturing processes, the normality assumption is often violated. Novelty detection more generally seeks abnormal (or novel) patterns in data, and novelty detection techniques can be applied to control charts in SPC. This dissertation consists of three components.;First, a bootstrap-based threshold for detecting abnormal patterns in multivariate T2 control chart is proposed. This approach can efficiently monitor a process when the distribution of observed data is nonnormal or unknown. The bootstrap method is a nonparametric technique that does not rely on the assumption of a parametric distribution of the observed data. Prior SPC literature only studies the bootstrap technique to develop univariate control charts to monitor a single process, while in this dissertation, the bootstrap technique is integrated with multivariate control charts.;Second, principal component analysis (PCA)-based control charts have been widely used to address problems posed by high correlations by reducing dimensionality. However, an assumption that the data are normally distributed has limited the use of PCA control charts. In this dissertation, the bootstrapping threshold approach and a kernel density estimation approach are employed for threshold development to yield nonparametric PCA control charts that do not require any distributional assumptions for their construction.;In novelty detection, support vector data description (SVDD) is a one-class classification technique that constructs a boundary in order to differentiate novel from normal patterns. However, boundaries constructed by SVDD do not consider the density of the data. Data points located in low density regions are more likely to be novel patterns because they are remote from their neighbors. This study presents a density-focused SVDD (DFSVDD), for which its boundary considers both shape and the dense region of the data.