No Aprendizado de Máquina Supervisionado---AM---o algoritmo de indução trabalha com um conjunto de exemplos de treinamento, no qual cada exemplo é constituído de um vetor com os valores dos atributos e as classes, e tem como tarefa induzir um classificador capaz de predizer a qual classe pertence um novo exemplo. Em geral, os algoritmos de indução baseiam-se nos exemplos de treinamento para a construção do classificador, sendo que uma representação inadequada desses exemplos, bem como inconsistências nos mesmos podem tornar a tarefa de aprendizado difícil. Um dos problemas centrais de AM é a Seleção de um Subconjunto de Atributos---SSA---cujo objetivo é diminuir o número de atributos utilizados na representação dos exemplos. São três as principais razões para a realização de SSA. A primeira razão é que a maioria dos algoritmos de AM, computacionalmente viáveis, não trabalham bem na presença de vários atributos. A segunda razão é que, com um número menor de atributos, o conceito induzido através do classificador pode ser melhor compreendido. E, a terceira razão é o alto custo para coletar e processar grande quantidade de informações. Basicamente, são três as abordagens para a SSA: embedded, filtro e wrapper. A Teoria de Rough Sets---RS---é uma abordagem matemática criada no início da década de 80, cuja principal funcionalidade são os redutos, e será tratada neste trabalho. Segundo essa abordagem, os redutos são subconjuntos mínimos de atributos que possuem a propriedade de preservar o poder de descrição do conceito relacionado ao conjunto de todos os atributos. Neste trabalho o enfoque esta na abordagem filtro para a realização da SSA utilizando como filtro os redutos calculados através de RS. São descritos vários experimentos sobre nove conjuntos de dados naturais utilizando redutos, bem como outros filtros para SSA. Feito isso, os atributos selecionados foram submetidos a dois algoritmos simbólicos de AM. Para cada conjunto de dados e indutor, foram realizadas várias medidas, tais como número de atributos selecionados, precisão e números de regras induzidas. Também, é descrito um estudo de caso sobre um conjunto de dados do mundo real proveniente da área médica. O objetivo desse estudo pode ser dividido em dois focos: comparar a precisão dos algoritmos de indução e avaliar o conhecimento extraído com a ajuda do especialista. Embora o conhecimento extraído não apresente surpresa, pôde-se confirmar algumas hipóteses feitas anteriormente pelo especialista utilizando outros métodos. Isso mostra que o Aprendizado de Máquina também pode ser visto como uma contribuição para outros campos científicos. / In Supervised Machine Learning---ML---an induction algorithm is typically presented with a set of training examples, where each example is described by a vector of feature values and a class label. The task of the induction algorithm is to induce a classifier that will be useful in classifying new cases. In general, the inductive-learning algorithms rely on existing provided data to build their classifiers. Inadequate representation of the examples through the description language as well as inconsistencies in the training examples can make the learning task hard. One of the main problems in ML is the Feature Subset Selection---FSS---problem, i.e. the learning algorithm is faced with the problem of selecting some subset of feature upon which to focus its attention, while ignoring the rest. There are three main reasons that justify doing FSS. The first reason is that most ML algorithms, that are computationally feasible, do not work well in the presence of many features. The second reason is that FSS may improve comprehensibility, when using less features to induce symbolic concepts. And, the third reason for doing FSS is the high cost in some domains for collecting data. Basically, there are three approaches in ML for FSS: embedded, filter and wrapper. The Rough Sets Theory---RS---is a mathematical approach developed in the early 1980\'s whose main functionality are the reducts, and will be treated in this work. According to this approach, the reducts are minimal subsets of features capable to preserve the same concept description related to the entire set of features. In this work we focus on the filter approach for FSS using as filter the reducts obtained through the RS approach. We describe a series of FSS experiments on nine natural datasets using RS reducts as well as other filters. Afterwards we submit the selected features to two symbolic ML algorithms. For each dataset, various measures are taken to compare inducers performance, such as number of selected features, accuracy and number of induced rules. We also present a case study on a real world dataset from the medical area. The aim of this case study is twofold: comparing the induction algorithms performance as well as evaluating the extracted knowledge with the aid of the specialist. Although the induced knowledge lacks surprising, it allows us to confirm some hypothesis already made by the specialist using other methods. This shows that Machine Learning can also be viewed as a contribution to other scientific fields.

Context: Software effort estimation is a very important component of the software development life cycle. It underpins activities such as planning, maintenance and bidding. Therefore, it has triggered much research over the past four decades, including many machine learning approaches. One popular approach, that has the benefit of accessible reasoning, is analogy-based estimation. Machine learning including analogy is known to significantly benefit from feature selection/weighting. Unfortunately feature weighting search is an NP hard problem, therefore computationally very demanding, if not intractable. Objective: Therefore, one objective of this research is to develop an effi cient and effective feature weighting algorithm for estimation by analogy. However, a major challenge for the effort estimation research community is that experimental results tend to be contradictory and also lack reliability. This has been paralleled by a recent awareness of how bias can impact research results. This is a contributory reason why software effort estimation is still an open problem. Consequently the second objective is to investigate research methods that might lead to more reliable results and focus on blinding methods to reduce researcher bias. Method: In order to build on the most promising feature weighting algorithms I conduct a systematic literature review. From this I develop a novel and e fficient feature weighting algorithm. This is experimentally evaluated, comparing three feature weighting approaches with a na ive benchmark using 2 industrial data sets. Using these experiments, I explore blind analysis as a technique to reduce bias. Results: The systematic literature review conducted identified 19 relevant primary studies. Results from the meta-analysis of selected studies using a one-sample sign test (p = 0.0003) shows a positive effect - to feature weighting in general compared with ordinary analogy-based estimation (ABE), that is, feature weighting is a worthwhile technique to improve ABE. Nevertheless the results remain imperfect so there is still much scope for improvement. My experience shows that blinding can be a relatively straightforward procedure. I also highlight various statistical analysis decisions which ought not be guided by the hunt for statistical significance and show that results can be inverted merely through a seemingly inconsequential statistical nicety. After analysing results from 483 software projects from two separate industrial data sets, I conclude that the proposed technique improves accuracy over the standard feature subset selection (FSS) and traditional case-based reasoning (CBR) when using pseudo time-series validation. Interestingly, there is no strong evidence for superior performance of the new technique when traditional validation techniques (jackknifing) are used but is more effi cient. Conclusion: There are two main findings: (i) Feature weighting techniques are promising for software effort estimation but they need to be tailored for target case for their potential to be adequately exploited. Despite the research findings showing that assuming weights differ in different parts of the instance space ('local' regions) may improve effort estimation results - majority of studies in software effort estimation (SEE) do not take this into consideration. This represents an improvement on other methods that do not take this into consideration. (ii) Whilst there are minor challenges and some limits to the degree of blinding possible, blind analysis is a very practical and an easy-to-implement method that supports more objective analysis of experimental results. Therefore I argue that blind analysis should be the norm for analysing software engineering experiments.

No description available.

Com o surgimento da tecnologia da informação, o processo de análise e interpretação de dados deixou de ser executado exclusivamente por seres humanos, passando a contar com auxílio computacional para a descoberta de conhecimento em grandes bancos de dados. Este auxílio exige uma organização e ordenação das atividades, antes manualmente exercidas, em um processo composto de três grandes etapas. A primeira etapa deste processo conta com uma tarefa de redução da dimensionalidade, que tem como objetivo a eliminação de atributos que não contribuem para a análise dos dados, resultando portanto, na seleção de um subconjunto dos atributos originais. A seleção de um subconjunto de atributos pode ser encarada como um problema de busca, já que há inúmeras possibilidades de combinação dos atributos originais em subconjuntos. Dessa forma, uma das estratégias de busca que pode ser adotada consiste na busca randômica, executada por um algoritmo genético ou pelas suas variações. Este trabalho propõe a aplicação de duas variações do algoritmo genético, Algoritmo Genético Construtivo e Algoritmo Genético Enviesado com Chave Aleatória, no problema de seleção de atributos em agrupamento de dados, já que estas duas variações ainda não foram aplicadas em tal problema. A fim de verificar o desempenho destas duas variações, comparou-se ambas com a abordagem tradicional do algoritmo genético. Efetuou-se também a comparação entre as duas variações. Para isto, foi utilizada três bases de dados retiradas do repositório UCI de aprendizado de máquinas. Os resultados obtidos mostraram que os desempenhos, em termos de qualidade da solução, dos algoritmos: genético construtivo e genético enviesado com chave aleatório foram melhores, de maneira geral, do que o desempenho da abordagem tradicional. Constatou-se também diferença significativa em termos de eficiência entre as duas variações e a abordagem tradicional. / With the advent of information technology, the process of analysis and interpretation of data left to be run exclusively by humans, going to rely on computational support for knowledge discovery in large databases. This aid requires an organization and sequencing of activities before manually performed in a compound of three major step process. The first step of this process has a reduced dimensionality task, which aims to eliminate attributes that do not contribute to the data analysis, resulting therefore, in selecting a subset of the original attributes. Selecting a subset of attributes can be viewed as a search problem, since there are numerous possible combinations of unique attributes into subsets. Thus, one search strategies that can be adopted is to randomly search, performed by a genetic algorithm or its variants. This paper proposes the application of two variations of the genetic algorithm, Constructive Genetic Algorithm and Biased Random Key Genetic Algorithm in the feature selection problem in data grouping, as these two variations have not been applied in such a problem. In order to verify the performance of the two variations, we compare them with the traditional algorithm, genetic algorithm. It was also executed the comparison between the two variations. For this, we used three databases removed from the UCI repository of machine learning. The results showed that the performance, in term of quality solution, of algorithms: genetic constructive and genetic biased with random key are better than the performance of the traditional approach. It was also observed a significant difference in efficiency between of the two variations and the traditional approach.

Com o surgimento da tecnologia da informação, o processo de análise e interpretação de dados deixou de ser executado exclusivamente por seres humanos, passando a contar com auxílio computacional para a descoberta de conhecimento em grandes bancos de dados. Este auxílio exige uma organização e ordenação das atividades, antes manualmente exercidas, em um processo composto de três grandes etapas. A primeira etapa deste processo conta com uma tarefa de redução da dimensionalidade, que tem como objetivo a eliminação de atributos que não contribuem para a análise dos dados, resultando portanto, na seleção de um subconjunto dos atributos originais. A seleção de um subconjunto de atributos pode ser encarada como um problema de busca, já que há inúmeras possibilidades de combinação dos atributos originais em subconjuntos. Dessa forma, uma das estratégias de busca que pode ser adotada consiste na busca randômica, executada por um algoritmo genético ou pelas suas variações. Este trabalho propõe a aplicação de duas variações do algoritmo genético, Algoritmo Genético Construtivo e Algoritmo Genético Enviesado com Chave Aleatória, no problema de seleção de atributos em agrupamento de dados, já que estas duas variações ainda não foram aplicadas em tal problema. A fim de verificar o desempenho destas duas variações, comparou-se ambas com a abordagem tradicional do algoritmo genético. Efetuou-se também a comparação entre as duas variações. Para isto, foi utilizada três bases de dados retiradas do repositório UCI de aprendizado de máquinas. Os resultados obtidos mostraram que os desempenhos, em termos de qualidade da solução, dos algoritmos: genético construtivo e genético enviesado com chave aleatório foram melhores, de maneira geral, do que o desempenho da abordagem tradicional. Constatou-se também diferença significativa em termos de eficiência entre as duas variações e a abordagem tradicional. / With the advent of information technology, the process of analysis and interpretation of data left to be run exclusively by humans, going to rely on computational support for knowledge discovery in large databases. This aid requires an organization and sequencing of activities before manually performed in a compound of three major step process. The first step of this process has a reduced dimensionality task, which aims to eliminate attributes that do not contribute to the data analysis, resulting therefore, in selecting a subset of the original attributes. Selecting a subset of attributes can be viewed as a search problem, since there are numerous possible combinations of unique attributes into subsets. Thus, one search strategies that can be adopted is to randomly search, performed by a genetic algorithm or its variants. This paper proposes the application of two variations of the genetic algorithm, Constructive Genetic Algorithm and Biased Random Key Genetic Algorithm in the feature selection problem in data grouping, as these two variations have not been applied in such a problem. In order to verify the performance of the two variations, we compare them with the traditional algorithm, genetic algorithm. It was also executed the comparison between the two variations. For this, we used three databases removed from the UCI repository of machine learning. The results showed that the performance, in term of quality solution, of algorithms: genetic constructive and genetic biased with random key are better than the performance of the traditional approach. It was also observed a significant difference in efficiency between of the two variations and the traditional approach.

This thesis deals with random-size subset selection and ranking procedures• • • )|(derived through likelihood ratios, mainly in terms of the P -approach.Let IT , . .. , IT, be k(> 2) populations such that IR.(i = l, . . . , k) hasJ_ K. — 12the normal distribution with unknwon mean 0. and variance a.a , where a.i i i2 . . is known and a may be unknown; and that a random sample of size n^ istaken from . To begin with, we give procedure (with tables) whichselects IT. if sup L(0;x) >c SUD L(0;X), where SÎ is the parameter space1for 0 = (0-^, 0^) ; where (with c: ß) is the set of all 0 with0. = max 0.; where L(*;x) is the likelihood function based on the total1sample; and where c is the largest constant that makes the rule satisfy theP*-condition. Then, we consider other likelihood ratios, with intuitivelyreasonable subspaces of ß, and derive several new rules. Comparisons amongsome of these rules and rule R of Gupta (1956, 1965) are made using differentcriteria; numerical for k=3, and a Monte-Carlo study for k=10.For the case when the populations have the uniform (0,0^) distributions,and we have unequal sample sizes, we consider selection for the populationwith min 0.. Comparisons with Barr and Rizvi (1966) are made. Generalizai<j<k Jtions are given.Rule R^ is generalized to densities satisfying some reasonable assumptions(mainly unimodality of the likelihood, and monotonicity of the likelihoodratio). An exponential class is considered, and the results are exemplifiedby the gamma density and the Laplace density. Extensions and generalizationsto cover the selection of the t best populations (using various requirements)are given. Finally, a discussion oil the complete ranking problem,and on the relation between subset selection based on likelihood ratios andstatistical inference under order restrictions, is given. / digitalisering@umu

For this project, we use a modified version of a previously developed mathematical model, which describes the relationships among matrix metalloproteinases (MMPs), their tissue inhibitors (TIMPs), and extracellular matrix (ECM). Our ultimate goal is to quantify and understand differences in parameter estimates between patients in order to predict future responses and individualize treatment for each patient. By analyzing parameter confidence intervals and confidence and prediction intervals for the state variables, we develop a parameter space reduction algorithm that results in better future response predictions for each individual patient. Moreover, use of another subset selection method, namely Structured Covariance Analysis, that considers identifiability of parameters, has been included in this work. Furthermore, to estimate parameters more efficiently and accurately, the standard error (SE- )optimal design method is employed, which calculates optimal observation times for clinical data to be collected. Finally, by combining different parameter subset selection methods and an optimal design problem, different cases for both finding optimal time points and intervals have been investigated.

In this thesis, inference of biological networks from in vivo data generated by perturbation experiments is considered, i.e. deduction of causal interactions that exist among the observed variables. Knowledge of such regulatory influences is essential in biology. A system property–interampatteness–is introduced that explains why the variation in existing gene expression data is concentrated to a few “characteristic modes” or “eigengenes”, and why previously inferred models have a large number of false positive and false negative links. An interampatte system is characterized by strong INTERactions enabling simultaneous AMPlification and ATTEnuation of different signals and we show that perturbation of individual state variables, e.g. genes, typically leads to ill-conditioned data with both characteristic and weak modes. The weak modes are typically dominated by measurement noise due to poor excitation and their existence hampers network reconstruction. The excitation problem is solved by iterative design of correlated multi-gene perturbation experiments that counteract the intrinsic signal attenuation of the system. The next perturbation should be designed such that the expected response practically spans an additional dimension of the state space. The proposed design is numerically demonstrated for the Snf1 signalling pathway in S. cerevisiae. The impact of unperturbed and unobserved latent state variables, that exist in any real biological system, on the inferred network and required set-up of the experiments for network inference is analysed. Their existence implies that a subnetwork of pseudo-direct causal regulatory influences, accounting for all environmental effects, in general is inferred. In principle, the number of latent states and different paths between the nodes of the network can be estimated, but their identity cannot be determined unless they are observed or perturbed directly. Network inference is recognized as a variable/model selection problem and solved by considering all possible models of a specified class that can explain the data at a desired significance level, and by classifying only the links present in all of these models as existing. As shown, these links can be determined without any parameter estimation by reformulating the variable selection problem as a robust rank problem. Solution of the rank problem enable assignment of confidence to individual interactions, without resorting to any approximation or asymptotic results. This is demonstrated by reverse engineering of the synthetic IRMA gene regulatory network from published data. A previously unknown activation of transcription of SWI5 by CBF1 in the IRMA strain of S. cerevisiae is proven to exist, which serves to illustrate that even the accumulated knowledge of well studied genes is incomplete. / Denna avhandling behandlar inferens av biologiskanätverk från in vivo data genererat genom störningsexperiment, d.v.s. bestämning av kausala kopplingar som existerar mellan de observerade variablerna. Kunskap om dessa regulatoriska influenser är väsentlig för biologisk förståelse. En system egenskap—förstärksvagning—introduceras. Denna förklarar varför variationen i existerande genexpressionsdata är koncentrerat till några få ”karakteristiska moder” eller ”egengener” och varför de modeller som konstruerats innan innehåller många falska positiva och falska negativa linkar. Ett system med förstärksvagning karakteriseras av starka kopplingar som möjliggör simultan FÖRSTÄRKning och förSVAGNING av olika signaler. Vi demonstrerar att störning av individuella tillståndsvariabler, t.ex. gener, typiskt leder till illakonditionerat data med både karakteristiska och svaga moder. De svaga moderna domineras typiskt av mätbrus p.g.a. dålig excitering och försvårar rekonstruktion av nätverket. Excitationsproblemet löses med iterativdesign av experiment där korrelerade störningar i multipla gener motverkar systemets inneboende försvagning av signaller. Följande störning bör designas så att det förväntade svaret praktiskt spänner ytterligare en dimension av tillståndsrummet. Den föreslagna designen demonstreras numeriskt för Snf1 signalleringsvägen i S. cerevisiae. Påverkan av ostörda och icke observerade latenta tillståndsvariabler, som existerar i varje verkligt biologiskt system, på konstruerade nätverk och planeringen av experiment för nätverksinferens analyseras. Existens av dessa tillståndsvariabler innebär att delnätverk med pseudo-direkta regulatoriska influenser, som kompenserar för miljöeffekter, generellt bestäms. I princip så kan antalet latenta tillstånd och alternativa vägar mellan noder i nätverket bestämmas, men deras identitet kan ej bestämmas om de inte direkt observeras eller störs. Nätverksinferens behandlas som ett variabel-/modelselektionsproblem och löses genom att undersöka alla modeller inom en vald klass som kan förklara datat på den önskade signifikansnivån, samt klassificera endast linkar som är närvarande i alla dessa modeller som existerande. Dessa linkar kan bestämmas utan estimering av parametrar genom att skriva om variabelselektionsproblemet som ett robustrangproblem. Lösning av rangproblemet möjliggör att statistisk konfidens kan tillskrivas individuella linkar utan approximationer eller asymptotiska betraktningar. Detta demonstreras genom rekonstruktion av det syntetiska IRMA genreglernätverket från publicerat data. En tidigare okänd aktivering av transkription av SWI5 av CBF1 i IRMA stammen av S. cerevisiae bevisas. Detta illustrerar att t.o.m. den ackumulerade kunskapen om välstuderade gener är ofullständig. / <p>QC 20130508</p>

DNA methylation is involved in various biological phenomena, and its dysregulation has been demonstrated as being correlated with a number of human disease processes, including cancers, autism, and autoimmune, mental health and neuro-degenerative ones. It has become important and useful in characterising and modelling these biological phenomena in or-der to understand the mechanism of such occurrences, in relation to both health and disease. An attempt has previously been made to map DNA methylation across human tissues, however, the means of distinguishing between methylated, unmethylated and differentially-methylated groups using DNA sequence features remains unclear. The aim of this study is therefore to: firstly, investigate DNA methylation classes and predict these based on DNA sequence features; secondly, to further identify methylation-associated DNA sequence features, and distinguish methylation differences between males and females in relation to both healthy and diseased, sta-tuses. This research is conducted in relation to three samples within nine biological feature sub-sets extracted from DNA sequence patterns (Human genome database). Two samples contain classes (methylated, unmethy-lated and differentially-methylated) within a total of 642 samples with 3,809 attributes driven from four human chromosomes, i.e. chromosomes 6, 20, 21 and 22, and the third sample contains all human chromosomes, which encompasses 1628 individuals, and then 1,505 CpG loci (features) were extracted by using Hierarchical clustering (a process Heatmap), along with pair correlation distance and then applied feature selection methods. From this analysis, author extract 47 features associated with gender and age, with 17 revealing significant methylation differences between males and females. Methylation classes prediction were applied a K-nearest Neighbour classifier, combined with a ten-fold cross- validation, since to some data were severely imbalanced (i.e., existed in sub-classes), and it has been established that direct analysis in machine-learning is biased towards the majority class. Hence, author propose a Modified- Leave-One-Out (MLOO) cross-validation and AdaBoost methods to tackle these issues, with the aim of compositing a balanced outcome and limiting the bias in-terference from inter-differences of the classes involved, which has provided potential predictive accuracies between 75% and 100%, based on the DNA sequence context.

Multi-armed bandit (MAB) algorithms could be used to select a subset of the k most informative summary statistics, from a pool of m possible summary statistics, by reformulating the subset selection problem as a MAB problem. This is suggested by experiments that tested five MAB algorithms (Direct, Halving, SAR, OCBA-m, and Racing) on the reformulated problem and comparing the results to two established subset selection algorithms (Minimizing Entropy and Approximate Sufficiency). The MAB algorithms yielded errors at par with the established methods, but in only a fraction of the time. Establishing MAB algorithms as a new standard for summary statistics subset selection could therefore save numerous scientists substantial amounts of time when selecting summary statistics for approximate bayesian computation.