Prediction and Privacy in Healthcare Analytics

Thesis (Master's)--University of Washington, 2016-03

A review of dynamic Bayesian network techniques with applications in healthcare risk modelling

Coping with an ageing population is a major concern for healthcare organisations around the world. The average cost of hospital care is higher than social care for older and terminally ill patients. Moreover, the average cost of social care increases with the age of the patient. Therefore, it is important to make efficient and fair capacity planning which also incorporates patient centred outcomes. Predictive models can provide predictions which their accuracy can be understood and quantified. Predictive modelling can help patients and carers to get the appropriate support services, and allow clinical decision-makers to improve care quality and reduce the cost of inappropriate hospital and Accident and Emergency admissions. The aim of this study is to provide a review of modelling techniques and frameworks for predictive risk modelling of patients in hospital, based on routinely collected data such as the Hospital Episode Statistics database. A number of sub-problems can be considered such as Length-of-Stay and End-of-Life predictive modelling. The methodologies in the literature are mainly focused on addressing the problems using regression methods and Markov models, and the majority lack generalisability. In some cases, the robustness, accuracy and re-usability of predictive risk models have been shown to be improved using Machine Learning methods. Dynamic Bayesian Network techniques can represent complex correlations models and include small probabilities into the solution. The main focus of this study is to provide a review of major time-varying Dynamic Bayesian Network techniques with applications in healthcare predictive risk modelling.

MASCEM - an electricity market simulator providing coalition support for virtual power players

This paper presents MASCEM - a multi-agent based electricity market simulator. MASCEM uses game theory, machine learning techniques, scenario analysis and optimization techniques to model market agents and to provide them with decision-support. This paper mainly focus on the MASCEM ability to provide the means to model and simulate Virtual Power Players (VPP). VPPs are represented as a coalition of agents, with specific characteristics and goals. The paper details some of the most important aspects considered in VPP formation and in the aggregation of new producers and includes a case study based on real data.

Incremental filter and wrapper approaches for feature discretization

Discrete data representations are necessary, or at least convenient, in many machine learning problems. While feature selection (FS) techniques aim at finding relevant subsets of features, the goal of feature discretization (FD) is to find concise (quantized) data representations, adequate for the learning task at hand. In this paper, we propose two incremental methods for FD. The first method belongs to the filter family, in which the quality of the discretization is assessed by a (supervised or unsupervised) relevance criterion. The second method is a wrapper, where discretized features are assessed using a classifier. Both methods can be coupled with any static (unsupervised or supervised) discretization procedure and can be used to perform FS as pre-processing or post-processing stages. The proposed methods attain efficient representations suitable for binary and multi-class problems with different types of data, being competitive with existing methods. Moreover, using well-known FS methods with the features discretized by our techniques leads to better accuracy than with the features discretized by other methods or with the original features. (C) 2013 Elsevier B.V. All rights reserved.

A Learning Algorithm and System Approach to Address Exceptional Events in Domestic Consumption Management

The integration of the Smart Grid concept into the electric grid brings to the need for an active participation of small and medium players. This active participation can be achieved using decentralized decisions, in which the end consumer can manage loads regarding the Smart Grid needs. The management of loads must handle the users’ preferences, wills and needs. However, the users’ preferences, wills and needs can suffer changes when faced with exceptional events. This paper proposes the integration of exceptional events into the SCADA House Intelligent Management (SHIM) system developed by the authors, to handle machine learning issues in the domestic consumption context. An illustrative application and learning case study is provided in this paper.

Analysis and visualization of energy use for university campus

The reduction of greenhouse gas emissions is one of the big global challenges for the next decades due to its severe impact on the atmosphere that leads to a change in the climate and other environmental factors. One of the main sources of greenhouse gas is energy consumption, therefore a number of initiatives and calls for awareness and sustainability in energy use are issued among different types of institutional and organizations. The European Council adopted in 2007 energy and climate change objectives for 20% improvement until 2020. All European countries are required to use energy with more efficiency. Several steps could be conducted for energy reduction: understanding the buildings behavior through time, revealing the factors that influence the consumption, applying the right measurement for reduction and sustainability, visualizing the hidden connection between our daily habits impacts on the natural world and promoting to more sustainable life. Researchers have suggested that feedback visualization can effectively encourage conservation with energy reduction rate of 18%. Furthermore, researchers have contributed to the identification process of a set of factors which are very likely to influence consumption. Such as occupancy level, occupants behavior, environmental conditions, building thermal envelope, climate zones, etc. Nowadays, the amount of energy consumption at the university campuses are huge and it needs great effort to meet the reduction requested by European Council as well as the cost reduction. Thus, the present study was performed on the university buildings as a use case to: a. Investigate the most dynamic influence factors on energy consumption in campus; b. Implement prediction model for electricity consumption using different techniques, such as the traditional regression way and the alternative machine learning techniques; and c. Assist energy management by providing a real time energy feedback and visualization in campus for more awareness and better decision making. This methodology is implemented to the use case of University Jaume I (UJI), located in Castellon, Spain.

Métodos de aprendizagem automática: um estudo baseado na avaliação e previsão de clientes bancários

Data Mining surge, hoje em dia, como uma ferramenta importante e crucial para o sucesso de um negócio. O considerável volume de dados que atualmente se encontra disponível, por si só, não traz valor acrescentado. No entanto, as ferramentas de Data Mining, capazes de transformar dados e mais dados em conhecimento, vêm colmatar esta lacuna, constituindo, assim, um trunfo que ninguém quer perder. O presente trabalho foca-se na utilização das técnicas de Data Mining no âmbito da atividade bancária, mais concretamente na sua atividade de telemarketing. Neste trabalho são aplicados catorze algoritmos a uma base de dados proveniente do call center de um banco português, resultante de uma campanha para a angariação de clientes para depósitos a prazo com taxas de juro favoráveis. Os catorze algoritmos aplicados no caso prático deste projeto podem ser agrupados em sete grupos: Árvores de Decisão, Redes Neuronais, Support Vector Machine, Voted Perceptron, métodos Ensemble, aprendizagem Bayesiana e Regressões. De forma a beneficiar, ainda mais, do que a área de Data Mining tem para oferecer, este trabalho incide ainda sobre o redimensionamento da base de dados em questão, através da aplicação de duas estratégias de seleção de atributos: Best First e Genetic Search. Um dos objetivos deste trabalho prende-se com a comparação dos resultados obtidos com os resultados presentes no estudo dos autores Sérgio Moro, Raul Laureano e Paulo Cortez (Sérgio Moro, Laureano, & Cortez, 2011). Adicionalmente, pretende-se identificar as variáveis mais relevantes aquando da identificação do potencial cliente deste produto financeiro. Como principais conclusões, depreende-se que os resultados obtidos são comparáveis com os resultados publicados pelos autores mencionados, sendo os mesmos de qualidade e consistentes. O algoritmo Bagging é o que apresenta melhores resultados e a variável referente à duração da chamada telefónica é a que mais influencia o sucesso de campanhas similares.

Multi-scale support vector algorithms for hot spot detection and modelling

The algorithmic approach to data modelling has developed rapidly these last years, in particular methods based on data mining and machine learning have been used in a growing number of applications. These methods follow a data-driven methodology, aiming at providing the best possible generalization and predictive abilities instead of concentrating on the properties of the data model. One of the most successful groups of such methods is known as Support Vector algorithms. Following the fruitful developments in applying Support Vector algorithms to spatial data, this paper introduces a new extension of the traditional support vector regression (SVR) algorithm. This extension allows for the simultaneous modelling of environmental data at several spatial scales. The joint influence of environmental processes presenting different patterns at different scales is here learned automatically from data, providing the optimum mixture of short and large-scale models. The method is adaptive to the spatial scale of the data. With this advantage, it can provide efficient means to model local anomalies that may typically arise in situations at an early phase of an environmental emergency. However, the proposed approach still requires some prior knowledge on the possible existence of such short-scale patterns. This is a possible limitation of the method for its implementation in early warning systems. The purpose of this paper is to present the multi-scale SVR model and to illustrate its use with an application to the mapping of Cs137 activity given the measurements taken in the region of Briansk following the Chernobyl accident.

Missing value imputation in longitudinal measures of alcohol consumption

Attrition in longitudinal studies can lead to biased results. The study is motivated by the unexpected observation that alcohol consumption decreased despite increased availability, which may be due to sample attrition of heavy drinkers. Several imputation methods have been proposed, but rarely compared in longitudinal studies of alcohol consumption. The imputation of consumption level measurements is computationally particularly challenging due to alcohol consumption being a semi-continuous variable (dichotomous drinking status and continuous volume among drinkers), and the non-normality of data in the continuous part. Data come from a longitudinal study in Denmark with four waves (2003-2006) and 1771 individuals at baseline. Five techniques for missing data are compared: Last value carried forward (LVCF) was used as a single, and Hotdeck, Heckman modelling, multivariate imputation by chained equations (MICE), and a Bayesian approach as multiple imputation methods. Predictive mean matching was used to account for non-normality, where instead of imputing regression estimates, "real" observed values from similar cases are imputed. Methods were also compared by means of a simulated dataset. The simulation showed that the Bayesian approach yielded the most unbiased estimates for imputation. The finding of no increase in consumption levels despite a higher availability remained unaltered. Copyright (C) 2011 John Wiley & Sons, Ltd.

Comparison of classification ability of hyperball algorithms to neural network and k-nearest neighbour algorithms

The main focus of this thesis is to evaluate and compare Hyperbalilearning algorithm (HBL) to other learning algorithms. In this work HBL is compared to feed forward artificial neural networks using back propagation learning, K-nearest neighbor and 103 algorithms. In order to evaluate the similarity of these algorithms, we carried out three experiments using nine benchmark data sets from UCI machine learning repository. The first experiment compares HBL to other algorithms when sample size of dataset is changing. The second experiment compares HBL to other algorithms when dimensionality of data changes. The last experiment compares HBL to other algorithms according to the level of agreement to data target values. Our observations in general showed, considering classification accuracy as a measure, HBL is performing as good as most ANn variants. Additionally, we also deduced that HBL.:s classification accuracy outperforms 103's and K-nearest neighbour's for the selected data sets.

Learning and Experiential Outcomes of Face-to-Face Versus Online Communications Courses

Higher education is rapidly trending toward the implementation of online (OL) courses and a blended facilitation style that incorporates both OL and face-to-face (FTF) classes. Though previous studies have explored the benefits and pitfalls of OL and blended learning formats from institutional, teacher, and student perspectives, scant research has examined learning outcomes for OL and FTF courses sharing identical content. This study used an explanatory mixed methods design—including pre- and post-test assessments, a questionnaire, and interviews—to explore similarities and differences in participant and teacher perceptions and outcomes (gain scores and final grades) of OL versus traditional FTF Communications courses, and to examine effects of students’ age and gender on learning preference and performance. Data collection occurred over a 4-month period and involved 183 student and 2 professor participants. The study used an SPSS program for data analysis and created a Microsoft Excel document to record themes derived from the questionnaire and interviews. Quantitative findings suggest there are no significant differences in gain scores, final grades, or other learning outcomes when comparing OL and FTF versions of identical Communications courses; however, qualitative findings indicate differences between facilitation styles based on student and professor perception. The study sheds light on student and faculty perceptions of facilitation styles and suggests areas for potential improvements in FTF- and OL-facilitated courses. The study ultimately recommends that students and faculty should have options when it comes to preferred delivery of course material.

Feature Selection and Classification Using Age Layered Population Structure Genetic Programming

The curse of dimensionality is a major problem in the fields of machine learning, data mining and knowledge discovery. Exhaustive search for the most optimal subset of relevant features from a high dimensional dataset is NP hard. Sub–optimal population based stochastic algorithms such as GP and GA are good choices for searching through large search spaces, and are usually more feasible than exhaustive and deterministic search algorithms. On the other hand, population based stochastic algorithms often suffer from premature convergence on mediocre sub–optimal solutions. The Age Layered Population Structure (ALPS) is a novel metaheuristic for overcoming the problem of premature convergence in evolutionary algorithms, and for improving search in the fitness landscape. The ALPS paradigm uses an age–measure to control breeding and competition between individuals in the population. This thesis uses a modification of the ALPS GP strategy called Feature Selection ALPS (FSALPS) for feature subset selection and classification of varied supervised learning tasks. FSALPS uses a novel frequency count system to rank features in the GP population based on evolved feature frequencies. The ranked features are translated into probabilities, which are used to control evolutionary processes such as terminal–symbol selection for the construction of GP trees/sub-trees. The FSALPS metaheuristic continuously refines the feature subset selection process whiles simultaneously evolving efficient classifiers through a non–converging evolutionary process that favors selection of features with high discrimination of class labels. We investigated and compared the performance of canonical GP, ALPS and FSALPS on high–dimensional benchmark classification datasets, including a hyperspectral image. Using Tukey’s HSD ANOVA test at a 95% confidence interval, ALPS and FSALPS dominated canonical GP in evolving smaller but efficient trees with less bloat expressions. FSALPS significantly outperformed canonical GP and ALPS and some reported feature selection strategies in related literature on dimensionality reduction.

Feature Selection and Classification Using Age Layered Population Structure Genetic Programming

The curse of dimensionality is a major problem in the fields of machine learning, data mining and knowledge discovery. Exhaustive search for the most optimal subset of relevant features from a high dimensional dataset is NP hard. Sub–optimal population based stochastic algorithms such as GP and GA are good choices for searching through large search spaces, and are usually more feasible than exhaustive and determinis- tic search algorithms. On the other hand, population based stochastic algorithms often suffer from premature convergence on mediocre sub–optimal solutions. The Age Layered Population Structure (ALPS) is a novel meta–heuristic for overcoming the problem of premature convergence in evolutionary algorithms, and for improving search in the fitness landscape. The ALPS paradigm uses an age–measure to control breeding and competition between individuals in the population. This thesis uses a modification of the ALPS GP strategy called Feature Selection ALPS (FSALPS) for feature subset selection and classification of varied supervised learning tasks. FSALPS uses a novel frequency count system to rank features in the GP population based on evolved feature frequencies. The ranked features are translated into probabilities, which are used to control evolutionary processes such as terminal–symbol selection for the construction of GP trees/sub-trees. The FSALPS meta–heuristic continuously refines the feature subset selection process whiles simultaneously evolving efficient classifiers through a non–converging evolutionary process that favors selection of features with high discrimination of class labels. We investigated and compared the performance of canonical GP, ALPS and FSALPS on high–dimensional benchmark classification datasets, including a hyperspectral image. Using Tukey’s HSD ANOVA test at a 95% confidence interval, ALPS and FSALPS dominated canonical GP in evolving smaller but efficient trees with less bloat expressions. FSALPS significantly outperformed canonical GP and ALPS and some reported feature selection strategies in related literature on dimensionality reduction.

GA approach for finding Rough Set decision rules based on bireducts

Feature selection plays an important role in knowledge discovery and data mining nowadays. In traditional rough set theory, feature selection using reduct - the minimal discerning set of attributes - is an important area. Nevertheless, the original definition of a reduct is restrictive, so in one of the previous research it was proposed to take into account not only the horizontal reduction of information by feature selection, but also a vertical reduction considering suitable subsets of the original set of objects. Following the work mentioned above, a new approach to generate bireducts using a multi--objective genetic algorithm was proposed. Although the genetic algorithms were used to calculate reduct in some previous works, we did not find any work where genetic algorithms were adopted to calculate bireducts. Compared to the works done before in this area, the proposed method has less randomness in generating bireducts. The genetic algorithm system estimated a quality of each bireduct by values of two objective functions as evolution progresses, so consequently a set of bireducts with optimized values of these objectives was obtained. Different fitness evaluation methods and genetic operators, such as crossover and mutation, were applied and the prediction accuracies were compared. Five datasets were used to test the proposed method and two datasets were used to perform a comparison study. Statistical analysis using the one-way ANOVA test was performed to determine the significant difference between the results. The experiment showed that the proposed method was able to reduce the number of bireducts necessary in order to receive a good prediction accuracy. Also, the influence of different genetic operators and fitness evaluation strategies on the prediction accuracy was analyzed. It was shown that the prediction accuracies of the proposed method are comparable with the best results in machine learning literature, and some of them outperformed it.

In silico analysis of mitochondrial proteins

Le rôle important joué par la mitochondrie dans la cellule eucaryote est admis depuis longtemps. Cependant, la composition exacte des mitochondries, ainsi que les processus biologiques qui sy déroulent restent encore largement inconnus. Deux facteurs principaux permettent dexpliquer pourquoi létude des mitochondries progresse si lentement : le manque defficacité des méthodes didentification des protéines mitochondriales et le manque de précision dans lannotation de ces protéines. En conséquence, nous avons développé un nouvel outil informatique, YimLoc, qui permet de prédire avec succès les protéines mitochondriales à partir des séquences génomiques. Cet outil intègre plusieurs indicateurs existants, et sa performance est supérieure à celle des indicateurs considérés individuellement. Nous avons analysé environ 60 génomes fongiques avec YimLoc afin de lever la controverse concernant la localisation de la bêta-oxydation dans ces organismes. Contrairement à ce qui était généralement admis, nos résultats montrent que la plupart des groupes de Fungi possèdent une bêta-oxydation mitochondriale. Ce travail met également en évidence la diversité des processus de bêta-oxydation chez les champignons, en corrélation avec leur utilisation des acides gras comme source dénergie et de carbone. De plus, nous avons étudié le composant clef de la voie de bêta-oxydation mitochondriale, lacyl-CoA déshydrogénase (ACAD), dans 250 espèces, couvrant les 3 domaines de la vie, en combinant la prédiction de la localisation subcellulaire avec la classification en sous-familles et linférence phylogénétique. Notre étude suggère que les gènes ACAD font partie dune ancienne famille qui a adopté des stratégies évolutionnaires innovatrices afin de générer un large ensemble denzymes susceptibles dutiliser la plupart des acides gras et des acides aminés. Finalement, afin de permettre la prédiction de protéines mitochondriales à partir de données autres que les séquences génomiques, nous avons développé le logiciel TESTLoc qui utilise comme données des Expressed Sequence Tags (ESTs). La performance de TESTLoc est significativement supérieure à celle de tout autre outil de prédiction connu. En plus de fournir deux nouveaux outils de prédiction de la localisation subcellulaire utilisant différents types de données, nos travaux démontrent comment lassociation de la prédiction de la localisation subcellulaire à dautres méthodes danalyse in silico permet daméliorer la connaissance des protéines mitochondriales. De plus, ces travaux proposent des hypothèses claires et faciles à vérifier par des expériences, ce qui présente un grand potentiel pour faire progresser nos connaissances des métabolismes mitochondriaux.