Muscle modelling allows us to create a synthetic muscle where we could simulate conditions in order to better understand the way a muscle works and reacts to inputs that would simply not be possible in a real person. Muscle modelling also allows us to simulate diseases and monitor their progression and behaviour with hopes of better understanding the disease, or creating tools that would improve the quality of life for many patients. Motor Neuron Diseases are a class of diseases where early intervention is crucial to any positive outcomes, by modelling this class of disease we can pave the way for better understanding, and early diagnosis. This thesis presents a Lower Motor Neurone Disease model that corresponds according to various measures to in-vivo observations described in the literature. Key findings of the experiments conducted, indicate that the order by which MUs die in MNDs significantly impacts how the muscle behaves.

The uniqueness of behavioral biometrics (e.g., voice or keystroke patterns) has been challenged by recent works. Statistical attacks have been proposed that infer general population statistics and target behavioral biometrics against a particular victim. We show that despite their success, these approaches require several attempts for successful attacks against different biometrics due to the different nature of overlap in users’ behavior for these biometrics. Furthermore, no mechanism has been proposed to date that detects statistical attacks. In this work, we propose a new hypervolumes-based statistical attack and show that unlike existing methods, it (1) is successful against a variety of biometrics, (2) is successful against more users, and (3) requires fewest attempts for successful attacks. More specifically, across five diverse biometrics, for the first attempt, on average our attack is 18 percentage points more successful than the second best (37% vs. 19%). Similarly, for the fifth attack attempt, on average our attack is 18 percentage points more successful than the second best (67% vs. 49%). We propose and evaluate a mechanism that can detect the more devastating statistical attacks. False rejects in biometric systems are common, and by distinguishing statistical attacks from false rejects, our defense improves usability and security. The evaluation of the proposed detection mechanism shows its ability to detect on average 94% of the tested statistical attacks with an average probability of 3% to detect false rejects as a statistical attack. Given the serious threat posed by statistical attacks to biometrics that are used today (e.g., voice), our work highlights the need for defending against these attacks.

NEFCLASS is a common example of a neuro-fuzzy system. The popular NEFCLASS classifier exhibits surprising behaviour when the feature values of the training and testing datasets exhibit significant skew. This paper presents a combined approach to improve the classification accuracy and interpretability of the NEFCLASS classifier, when data distribution exhibits positive skewness. The proposed model consists of two steps. Firstly, we used an alternative discretization method to initialize fuzzy sets. Secondly, we devised a statistical rule pruning algorithm based on adjusted residual to reduce the number of rules, thus improving interpretability. This method improves the interpretability of NEFCLASS without significant accuracy deterioration. Moreover, a hybrid approach combining the two approaches is developed to increase the accuracy-interpretability trade-off of NEFCLASS.

This paper evaluates the classification of multisample problems, such as electromyographic (EMG) data, by making aggregate features available to a per-sample classifier. It is found that the accuracy of this approach is superior to that of traditional methods such as majority vote for this problem. The classification improvements of this method, in conjunction with a confidence measure expressing the per-sample probability of classification failure (i.e., a hazard function) is described and measured. Results are expected to be of interest in clinical decision support system development.

Rule-based classifiers allow rationalization of classifications made. This in turn improves understanding which is essential for effective decision support. As a rule based classifier, the pattern discovery (PD) algorithm functions well in discrete, nominal and continuous data domains. A drawback when using PD as a classifier for decision support is that it has an unbounded decision space that confounds the understanding of the degree of support for a decision. Incorporating PD into a fuzzy inference system (FIS) allows the the degree of support for a decision to be expressed with intuitively understandable terms. In addition, using discrete algorithms in continuous domains can result in reduced accuracy due to quantization. Fuzzification reduces this ldquocost of quantizationrdquo and improves classification performance. In this work, the PD algorithm was used as a source of rules for a series of FISs implemented using different rule weighting and defuzzification schemes, each providing a linguistic basis for rule description and a bounded space for expression of decision support. The output of each FIS consists of a suggested outcome, a strong confidence metric describing suggestions within this space and a linguistic expression of the rules. This constitutes a stronger basis for decision making than that provided by PD alone. A variety of synthetic, continuous class distributions with varying degrees of separation was used to evaluate the performance of fuzzy, PD, back-propagation and Bayesian classifiers. Overall, the accuracy of the fuzzy system was found to be similar, but slightly below, that of the inherently continuous valued classifiers and was somewhat improved with respect to the PD classifiers. For the difficult spiral class distributions studied, the fuzzy classifiers were able to make more classifications than the PD classifiers. The correct classification rates for the fuzzy classifiers were similar across the various rule weighting and defuzzification schemes, demonstrating the strength of the statistical method for rule generation. Analysis of several real-world data sets shows that a PD-based FIS has comparable performance to a neuro-fuzzy system. The use of a PD based FIS however, provides insight into the structure of the data analyzed not available through the other approaches.

This work presents a system intended to assist in the diagnosis of muscular disease based on the decomposition of electromyographic data through quantitative means. The performance of the system is evaluated both through quantitative analysis of correct suggestions as well as descriptively through an examination of the decision space presented in several pertinent examples. Data used for this analysis is synthetically created gold-standard data produced through a previously described muscle simulator and prepared through application of the DQEMG program. The design of the decision support system is based on a conceptual model reflecting the decision process present in clinical work; the provision of a drill-down exploration mechanism to allow investigation of the statistical support underlying the decision process allows access to the needed statistical confidences required for informed clinical diagnosis while providing summary data display to allow a clinician access to muscle-level characterizations in a reproducible and quantifiable way. The use of the "pattern discovery" algorithm provides a statistically sound mechanism for the generation of rules from a training data set. The exceptional performance observed here provides confidence in the system as a potential tool for use in clinical diagnosis

A framework for the development of a decision support system (DSS) that exhibits uncommonly transparent rule-based inference logic is introduced. A DSS is constructed by marrying a statistically based fuzzy inference system (FIS) with a user interface, allowing drill-down exploration of the underlying statistical support, providing transparent access to both the rule-based inference as well as the underlying statistical basis for the rules. The FIS is constructed through a "pattern discovery" based analysis of training data. Such an analysis yields a rule base characterized by simple explanations for any rule or data division in the extracted knowledge base. The reliability of a fuzzy inference is well predicted by a confidence measure that determines the probability of a correct suggestion by examination of values produced within the inference calculation. The combination of these components provides a means of constructing decision support systems that exhibit a degree of transparency beyond that commonly observed in supervised-learning-based methods. A prototype DSS is analyzed in terms of its workflow and usability, outlining the insight derived through use of the framework. This is demonstrated by considering a simple synthetic data example and a more interesting real-world example application with the goal of characterizing patients with respect to risk of heart disease. Specific input data samples and corresponding output suggestions created by the system are presented and discussed. The means by which the suggestions made by the system may be used in a larger decision context is evaluated

The pattern discovery (PD) algorithm of Wang and Wong was applied as a classifier to several continuous-valued data sets generated to explore performance across a selection of interesting linearly and non-linearly separable class distributions. Performance of several configurations of PD and backpropagation (BP) neural network classifiers and a minimum inter-class distance (MICD) classifier was quantified and compared. The best performance of the PD and BP classifiers were found to be similar for all class distributions studied and close to the optimal IMICD performance for linearly separable class distributions. The performance of both PD and BP classifiers was dependent on the classifier configuration. PD classifier performance depended on the number of intervals used to quantize the continuous data in a predictable, class-distribution independent way. BP performance depended on the number of hidden nodes in a way which was class-distribution dependent and difficult to determine a priori. The transparency and statistical validity of the patterns used and the decisions made by PD classifiers make them highly suitable for problems in which the rationale and confidence of classifications are required so that multiple classifications can be effectively combined to support decisions in a broader context such as medical diagnosis. The strong absolute and relative performance of PD classifiers and the relative simplicity of their implementation when applied to continuous-valued data suggest that they can be effectively utilized in decision support systems in which the underlying data is continuous or discrete valued.

Many genetic algorithms have complex fitness functions which can easily be calculated in parallel, given the tools to do so. This paper explores the use of a tool to gather spare computing cycles from a variable set of machines to allow convergence of GAs of this type. A modification to the steady-state model for GAs allows us to use the fault-prone behavior of an underlying thin networked computation system as noise within the GA itself. This "real" noise is incorporated into the GA, maintaining the drive towards convergence in the case of the heavily noisy network environment.