Background: The main cause of diabetes is due to high sugar levels in the blood. There is no permanent cure for diabetes. However, it can be prevented by early diagnosis. In recent years, the hype for Machine Learning is increasing in disease prediction especially during COVID-19 times. In the present scenario, it is difficult for patients to visit doctors. A possible framework is provided using Machine Learning which can detect diabetes at early stages. Objectives: This thesis aims to identify the critical features that impact gestational (Type-3) diabetes and experiments are performed to identify the efficient algorithm for Type-3 diabetes prediction. The selected algorithms are Decision Trees, RandomForest, Support Vector Machine, Gaussian Naive Bayes, Bernoulli Naive Bayes, Laplacian Support Vector Machine. The algorithms are compared based on the performance. Methods: The method consists of gathering the dataset and preprocessing the data. SelectKBestunivariate feature selection was performed for selecting the important features, which influence the Type-3 diabetes prediction. A new dataset was created by binning some of the important features from the original dataset, leading to two datasets, non-binned and binned datasets. The original dataset was imbalanced due to the unequal distribution of class labels. The train-test split was performed on both datasets. Therefore, the oversampling technique was performed on both training datasets to overcome the imbalance nature. The selected Machine Learning algorithms were trained. Predictions were made on the test data. Hyperparameter tuning was performed on all algorithms to improve the performance. Predictions were made again on the test data and accuracy, precision, recall, and f1-score were measured on both binned and non-binned datasets. Results: Among selected Machine Learning algorithms, Laplacian Support Vector Machineattained higher performance with 89.61% and 86.93% on non-binned and binned datasets respectively. Hence, it is an efficient algorithm for Type-3 diabetes prediction. The second best algorithm is Random Forest with 74.5% and 72.72% on non-binned and binned datasets. The non-binned dataset performed well for the majority of selected algorithms. Conclusions: Laplacian Support Vector Machine scored high performance among the other algorithms on both binned and non-binned datasets. The non-binned dataset showed the best performance in almost all Machine Learning algorithms except Bernoulli naive Bayes. Therefore, the non-binned dataset is more suitable for the Type-3 diabetes prediction.

In 2018, Europol (2018) estimated that more than $22 billion dollars were laundered in Europe by using cryptocurrencies. The Financial Action Task Force explains that moneylaunderers may exchange their illicitly gained fiat-money for crypto, launder that crypto by distributing the funds to multiple accounts and then re-exchange the crypto back to fiat-currency. This process of exchanging currencies is done through a cryptocurrency exchange, giving the exchange an ideal position to prevent money laundering from happening as it acts as middleman (FATF, 2021). However, current AML efforts at these exchanges have shown to be outdated and need to be improved. Furthermore, Weber et al. (2019) argue that machine learning could be used for this endeavor. The study's purpose is to investigate how machine learning can be used to combat money laundering activities performed using cryptocurrency. This is done by exploring what machine learning algorithms are suitable for this purpose. In addition, the study further seeks to understand the applicability of the investigated algorithms by exploring their fit at cryptocurrency exchanges. To answer the research question, four supervised-learning algorithms are compared by using the Bitcoin Elliptic Dataset. Moreover, with the objective of quantitively understanding the algorithmic performance differences, three key evaluation metrics are used: F1-score, precision and recall. Then, in order to understand the investigated algorithms applicability, two complementary qualitative interviews are performed at Swedish cryptocurrency exchanges. The study cannot conclude if there is a most suitable algorithm for detecting transactions related to money-laundering. However, the applicability of the decision tree algorithm seems to be more promising at Swedish cryptocurrency exchanges, compared to the other three algorithms. / Europol (2018) uppskattade år 2018, att mer än 22 miljarder USD tvättades i Europa genom användning av kryptovalutor. Financial Action Task Force förklarar att penningtvättare kan byta deras olagligt förvärvade fiat-valutor mot kryptovaluta, tvätta kryptovalutan genom att fördela tillgångarna till ett flertal konton och sedan återväxla kryptovalutan tillbaka till fiat-valuta. Denna process, att växla valutor, görs genom en kryptovalutaväxlare, vilket ger växlaren en ideal position för att förhindra att tvättning sker eftersom de agerar som mellanhänder (FATF, 2021). Dock har de aktuella AMLansträngningarna vid dessa växlare visat sig vara föråldrade och i behov av förbättring. Dessutom hävdar Weber et al. (2019) att maskininlärning skulle kunna användas i denna strävan. Denna studies syfte är att undersöka hur maskininlärning kan användas för att bekämpa penningtvättaktiviteter där kryptovaluta används. Detta görs genom att utforska vilka maskininlärningsalgoritmer som är användbara för detta ändamål. Dessutom strävar undersökningen till att ge förståelse för tillämpligheten hos de undersökta algoritmerna genom att utforska deras lämplighet hos kryptovalutaväxlare. För att besvara frågeställningen har fyra supervised-learning algoritmer jämförts genom att använda Bitcoin Elliptic Dataset. För att kvantitativt förstå olikheterna i algoritmisk prestanda, har tre utvärderingsverktyg använts: F1-score, Precision och Recall. Slutligen, för att ytterligare förstå de undersökta algoritmernas tillämplighet, har två kompletterande kvalitativa intervjuer med svenska kryptovalutaväxlare gjorts. Studien kan inte dra slutsatsen att det finns en bästa algoritm för att upptäcka transaktioner som kan relateras till penningtvätt. Dock verkar tillämpbarheten hos decision tree algoritmen vara mer lovande vid de svenska kyptovalutaväxlarna än de tre andra algoritmerna.

Contexte : L’activité physique est une composante centrale du développement physique, psychologique et social de l'enfant, particulièrement au sein d'une société où l'impact de la sédentarité et de l'obésité devient de plus en plus important. Cependant, les trajectoires d’activité physique hors école et leurs déterminants sont peu étudiés et les connaissances sur ce sujet sont limitées. Il est également notoire que les types d’activité physique sont rarement pris en considération. Objectif : Ce mémoire a pour but (a) de déterminer les trajectoires de pratique d’activité physique au cours du développement des enfants (b) de valider l’association entre l’activité physique supervisée et l’activité non supervisée et (c) d’identifier les déterminants au niveau du quartier, de la famille et des caractéristiques individuelles associés aux trajectoires de pratique d’activité physique supervisée et non supervisée. Participants : 1 814 enfants (51% garçons) nés en 1998 ayant participé à l’Étude Longitudinale du Développement des Enfants du Québec (ELDEQ). Les données récoltées proviennent uniquement de leur mère. Mesures : La fréquence de l’activité physique supervisée et non supervisée a été mesurée à quatre reprises alors que les enfants étaient âgés entre 5 et 8 ans. Les déterminants ainsi que les variables contrôles ont été mesurés alors que les enfants avaient 4 ou 5 ans. Résultats : Trois trajectoires d’activité physique supervisée et non supervisée ont été identifiées. Les résultats suggèrent que les trajectoires d’activité physique supervisée, représentant respectivement 10%, 55.3% et 34.7% de la population, sont relativement stables même si elles subissent une légère augmentation avec le temps. Des trois trajectoires d’activité physique non supervisée représentant respectivement 14.1%, 28.1% et 57.8% de la population, une augmente considérablement avec le temps alors iv que les deux autres sont stables. Ces deux séries de trajectoires ne sont pas associées significativement entre elles. L’éducation de la mère, l’entraide dans le quartier de résidence ainsi que la prosocialité des enfants déterminent les deux types d’activité physique. La suffisance de revenu et la pratique sportive de la mère sont associées seulement aux trajectoires d’activité physique supervisée. La famille intacte discrimine l’appartenance aux trajectoires d’activité physique non supervisée. Conclusion : Premièrement, la pratique de l’activité physique est relativement stable entre 5 et 8 ans. Deuxièmement, l’activité physique supervisée ainsi que l’activité physique non supervisée sont deux pratiques qui se développent différemment et qui possèdent leurs propres déterminants. Troisièmement, une approche écologique permet de mieux saisir la complexité de ces deux processus. / Context : Physical activity is a central component of a child physical, psychological and social development, most importantly in a society where sedentary behaviors and obesity become a more significant problematic. Few studies have investigated the developmental trajectories and predictors of physical activity over time. Furthermore, even fewer studies have investigated supervised and non-supervised physical activity separately. Objectives : The present study has for main goals (a) to identify developmental trajectories of supervised and non-supervised physical activity in elementary school children (b) to assess the link between these two types of physical activity (c) to identify neighborhood, family and individual predictors of these two types of physical activity. Participants: 1 814 children (51% boys) born in 1998 who participated in the Quebec Longitudinal Study of Child Development (QLSCD). Data were mainly collected through mothers’ report. Measures : The frequency of physical activity was measured at four time points when children were aged between 5 and 8 years old. Predictors and control variables were assessed when children were 4 or 5 years old. Results : Three trajectories of supervised and non-supervised activities have been identified. Trajectories of supervised physical activity (10%, 55.3% et 34.7%) are relatively stable although they are slightly increasing over time. Trajectories of nonsupervised physical activity (14.1%, 28.1% et 57.8%) are relatively stable although one group (28.1%) is increasing considerably. Supervised and non-supervised physical activity trajectories are not related to each other. Mother’s education, neighborhood safety and child’s prosociality are related to high frequency of both physical activities. Sufficient revenue and mother’s involvement in sport is related to frequent supervised vi physical activity trajectories as intact family predict less frequent non-supervised physical activity trajectories. Conclusion : First, involvement in supervised and non-supervised physical activity is relatively stable between 5 and 8 years old. Second, supervised and non-supervised physical activity appear to be two different processes that have their own set of predictors. Third, an ecological and multidimensional approach is required to capture the complexity of these two processes.

Introdução: Quedas representam risco extremamente incidente entre idosos, e sua recuperação produz altos custos. Algumas das causas mais comuns podem ser atenuadas por exercícios, se oferecidos de forma acessível. Objetivos: Comparar os efeitos de um treinamento físico multimodal quando realizado de forma supervisionada e semissupervisionada, sobre variáveis reconhecidamente relacionadas ao risco de quedas em idoso com preservada independência e histórico de quedas. Métodos: Setenta e seis idosos com histórico de quedas, acima de 70 anos, média de 79,06 anos (±4,55), foram avaliados sobre a saúde geral, histórico e risco de quedas, perigos domésticos, e foram submetidos aos seguintes testes: Timed up and Go (TUG), Walk Performance Test (WPT), Berg Balance Scale (BBS), avaliação isocinética do joelho e os seguintes testes em plataforma de equilibrio: Tandem Walk (TW), Sit to Stand (STS), Step up Over (SUO), Limits of Stability (LOS) e Modified Clinical Test of Sensory Integration on Balance (MCTSIB). Foram aleatoriamente alocados em 3 grupos: Supervisionado (S), orientado em todas as sessões, Semissupervisionado (SS), orientado quinzenalmente a executar exercícios em casa, e Controle (C), sem intervenção. O programa de exercícios multimodais foi executado em 3 sessões semanais de 50 minutos, por 4 meses. Participantes registraram quedas em calendário, e avaliações foram repetidas ao final do período. Resultados: Após intervenção o grupo S reduziu tempo do TUG (p<0,001) e no WPT (p<0,001) e aumentou a pontuação do BBS (p= 0,018), a Potência Média (p<0,001), o Pico de Torque/ Peso (p= 0,036) e a Média do Pico de Torque (p= 0,006) na flexão direita. Reduziu Tempo de Transferência no STS (p= 0,039), o Índice de Impacto na descida no SUO (p= 0.047), e a Oscilação no MCTSIB na 1ª (p= 0,037) e na 4ª (p= 0,032) condições avaliadas. No LOS, aumentou Velocidade (p<0,001), a Máxima Excursão (p<0,001) e o Controle de Direção (p= 0,004). O grupo SS reduziu o tempo no TUG (p= 0,001), aumentou o Índice de Fadiga na flexão do joelho direito (p= 0,043), aumentou Velocidade e reduziu Oscilação no TW (p= 0,008 e 0,020 respectivamente). No LOS, aumentou Velocidade (p= 0,023), a Máxima Excursão (p= 0,035) e o Controle de Direção (p= 0,006). O grupo C reduziu Velocidade no TW (p= 0,033) e aumentou o Índice de Fadiga na flexão direita (p= 0,017). O grupo S apresentou magnitude do efeito diferente na Potência Média da Flexão do Joelho direito (p= 0,002 para S versus SS, e p= 0,004 para S versus C). Os grupos S e C apresentaram diferença entre si na variação da Velocidade do LOS (p= 0,003). Os grupos S e SS obtiveram alterações diferentes do grupo C no TUG (p= 0,003 para C vs. S, e p= 0,021 para C vs. SS), e na Velocidade do TW (p= 0,007 para C vs. S, e p= 0,003 para C vs. SS). Conclusões: Numa população de idosos não institucionalizados, com independência preservada, baixa renda, pouca escolaridade, e com histórico de quedas, um treinamento físico multimodal, aplicado tanto de forma semissupervisionada, em casa, quanto de forma supervisionada, no centro de saúde, pode ser efetivo em melhorar variáveis previamente reconhecidas como sendo altamente relacionadas ao risco de quedas. Os resultados equivalentes entre os grupos S e SS impedem-nos de afirmar que a supervisão acrescente expressiva extensão a este benefício / Background: Falls are an extremely incidental healthcare risk among the geriatric populations and lead to high recuperative costs. Muscle weakness and balance impairment are among the most common causes and can be attenuated by exercises, if provided in an accessible way. Objectives: To compare the effects on variables related to falls risk, of a fully supervised center-based and a semi-supervised home-based multi-modal exercise program in elderly with preserved independence, and history of falls. Methods: Seventy six older adults with history of falls, over 70 years old, mean age of 79.06 years (±4.55) were assessed about general health, falls history and risk, home hazard and were submitted to the following tests: Timed up and Go (TUG), Walk Performance Test (WPT), Berg Balance Scale (BBS), Knee Isokinetic dynamometer test, and five tests on balance force plate: Tandem Walk (TW), Sit to Stand (STS), Step up Over (SUO), Limits of Stability (LOS) and modified Clinical Test of Sensory Integration on Balance (MCTSIB). Participants were randomized into three groups: supervised (S) that was instructed in all sessions, semi-supervised (SS) that received orientation every other week and performed the exercises at home, and control (C) that did not receive any exercise intervention. The multi-modal program consisted in three 50-minute sessions per week over four months. Participants recorded falls in a calendar and assessments were repeated at the end of the period Results: After intervention, S groups reduced time in TUG (p<0.001) and WPT (p<0.001), increased the BBS score (p= 0.018), the Average Power (p<0.001), the Peak Torque/Weight (p= 0.036), and the average Peak Torque (p= 0.006) of right knee flexion. It reduced Transfer Time in STS (p= 0.039), o Impact Index on SUO (p= 0.047), and End Sway on MCTSIB on 1st (p= 0.037) and 4th (p= 0.032) conditions assessed. On LOS, increased Movement Velocity (p<0.001), Maximum Excursion (p<0.001), and Directional Control (p= 0.004). The SS group reduced TUG (p= 0.001), increased Fatigue Work on right knee flexion (p= 0.043), increased Speed and reduced End Sway on TW (p= 0.008 e 0.020 respectively). On LOS, increased the velocity (p= 0,023), the Maximum Excursion (p= 0.035) and Directional Control (p= 0.006). The C group reduced TW speed (p= 0.033) and increased Fatigue Work of right knee flexion (p= 0.017). The S group showed different magnitude of effect in Average Power of right knee flexion (p= 0.002 for S vs. SS, and p= 0,004 for S vs. C). Groups S and C were different from each other on LOS Velocity (p= 0.003). Comparing to C, both trained groups, S and SS, had different magnitude of effect on TUG (p= 0.003 for C vs. S, and p= 0.021 for C vs. SS), and TW Speed (p= 0.007 for C vs. S, and p= 0.003 for C vs. SS). Conclusions: In a community-dwelling elderly population with preserved independence, low income and minimal education, with history of falls, a semi-supervised home-based and supervised center-based multi-modal exercise program, may be effective in improving variables previously recognized as highly related to falls risk. The similar results between trained groups prevent us to affirm that supervision adds expressive extent to the benefit

Supervised learning tasks like building a classifier, estimating the error rate of the predictors, are typically performed with labeled data. In most cases, obtaining labeled data is costly as it requires manual labeling. On the other hand, unlabeled data is available in abundance. In this thesis, we discuss methods to perform supervised learning tasks with no labeled data. We prove consistency of the proposed methods and demonstrate its applicability with synthetic and real world experiments. In some cases, small quantities of labeled data maybe easily available and supplemented with large quantities of unlabeled data (semi-supervised learning). We derive the asymptotic efficiency of generative models for semi-supervised learning and quantify the effect of labeled and unlabeled data on the quality of the estimate. Another independent track of the thesis is efficient computational methods for nonnegative tensor factorization (NTF). NTF provides the user with rich modeling capabilities but it comes with an added computational cost. We provide a fast algorithm for performing NTF using a modified active set method called block principle pivoting method and demonstrate its applicability to social network analysis and text mining.

Hurst exponent and variance are two quantities that often characterize real-life, highfrequency observations. We develop the method for simultaneous estimation of a timechanging Hurst exponent H(t) and constant scale (variance) parameter C in a multifractional Brownian motion model in the presence of white noise based on the asymptotic behavior of the local variation of its sample paths. We also discuss the accuracy of the stable and simultaneous estimator compared with a few selected methods and the stability of computations that use adapted wavelet filters. Multifractals have become popular as flexible models in modeling real-life data of high frequency. We developed a method of testing whether the data of high frequency is consistent with monofractality using meaningful descriptors coming from a wavelet-generated multifractal spectrum. We discuss theoretical properties of the descriptors, their computational implementation, the use in data mining, and the effectiveness in the context of simulations, an application in turbulence, and analysis of coding/noncoding regions in DNA sequences. The wavelet thresholding is a simple and effective operation in wavelet domains that selects the subset of wavelet coefficients from a noised signal. We propose the selection of this subset in a semi-supervised fashion, in which a neighbor structure and classification function appropriate for wavelet domains are utilized. The decision to include an unlabeled coefficient in the model depends not only on its magnitude but also on the labeled and unlabeled coefficients from its neighborhood. The theoretical properties of the method are discussed and its performance is demonstrated on simulated examples.

Gaussian Processes (GPs) are promising Bayesian methods for classification and regression problems. They have also been used for semi-supervised classification tasks. In this thesis, we propose new algorithms for solving semi-supervised binary classification problem using GP regression (GPR) models. The algorithms are closely related to semi-supervised classification based on support vector regression (SVR) and maximum margin clustering. The proposed algorithms are simple and easy to implement. Also, the hyper-parameters are estimated without resorting to expensive cross-validation technique. The algorithm based on sparse GPR model gives a sparse solution directly unlike the SVR based algorithm. Use of sparse GPR model helps in making the proposed algorithm scalable. The results of experiments on synthetic and real-world datasets demonstrate the efficacy of proposed sparse GP based algorithm for semi-supervised classification.

Contexte : L’activité physique est une composante centrale du développement physique, psychologique et social de l'enfant, particulièrement au sein d'une société où l'impact de la sédentarité et de l'obésité devient de plus en plus important. Cependant, les trajectoires d’activité physique hors école et leurs déterminants sont peu étudiés et les connaissances sur ce sujet sont limitées. Il est également notoire que les types d’activité physique sont rarement pris en considération. Objectif : Ce mémoire a pour but (a) de déterminer les trajectoires de pratique d’activité physique au cours du développement des enfants (b) de valider l’association entre l’activité physique supervisée et l’activité non supervisée et (c) d’identifier les déterminants au niveau du quartier, de la famille et des caractéristiques individuelles associés aux trajectoires de pratique d’activité physique supervisée et non supervisée. Participants : 1 814 enfants (51% garçons) nés en 1998 ayant participé à l’Étude Longitudinale du Développement des Enfants du Québec (ELDEQ). Les données récoltées proviennent uniquement de leur mère. Mesures : La fréquence de l’activité physique supervisée et non supervisée a été mesurée à quatre reprises alors que les enfants étaient âgés entre 5 et 8 ans. Les déterminants ainsi que les variables contrôles ont été mesurés alors que les enfants avaient 4 ou 5 ans. Résultats : Trois trajectoires d’activité physique supervisée et non supervisée ont été identifiées. Les résultats suggèrent que les trajectoires d’activité physique supervisée, représentant respectivement 10%, 55.3% et 34.7% de la population, sont relativement stables même si elles subissent une légère augmentation avec le temps. Des trois trajectoires d’activité physique non supervisée représentant respectivement 14.1%, 28.1% et 57.8% de la population, une augmente considérablement avec le temps alors iv que les deux autres sont stables. Ces deux séries de trajectoires ne sont pas associées significativement entre elles. L’éducation de la mère, l’entraide dans le quartier de résidence ainsi que la prosocialité des enfants déterminent les deux types d’activité physique. La suffisance de revenu et la pratique sportive de la mère sont associées seulement aux trajectoires d’activité physique supervisée. La famille intacte discrimine l’appartenance aux trajectoires d’activité physique non supervisée. Conclusion : Premièrement, la pratique de l’activité physique est relativement stable entre 5 et 8 ans. Deuxièmement, l’activité physique supervisée ainsi que l’activité physique non supervisée sont deux pratiques qui se développent différemment et qui possèdent leurs propres déterminants. Troisièmement, une approche écologique permet de mieux saisir la complexité de ces deux processus. / Context : Physical activity is a central component of a child physical, psychological and social development, most importantly in a society where sedentary behaviors and obesity become a more significant problematic. Few studies have investigated the developmental trajectories and predictors of physical activity over time. Furthermore, even fewer studies have investigated supervised and non-supervised physical activity separately. Objectives : The present study has for main goals (a) to identify developmental trajectories of supervised and non-supervised physical activity in elementary school children (b) to assess the link between these two types of physical activity (c) to identify neighborhood, family and individual predictors of these two types of physical activity. Participants: 1 814 children (51% boys) born in 1998 who participated in the Quebec Longitudinal Study of Child Development (QLSCD). Data were mainly collected through mothers’ report. Measures : The frequency of physical activity was measured at four time points when children were aged between 5 and 8 years old. Predictors and control variables were assessed when children were 4 or 5 years old. Results : Three trajectories of supervised and non-supervised activities have been identified. Trajectories of supervised physical activity (10%, 55.3% et 34.7%) are relatively stable although they are slightly increasing over time. Trajectories of nonsupervised physical activity (14.1%, 28.1% et 57.8%) are relatively stable although one group (28.1%) is increasing considerably. Supervised and non-supervised physical activity trajectories are not related to each other. Mother’s education, neighborhood safety and child’s prosociality are related to high frequency of both physical activities. Sufficient revenue and mother’s involvement in sport is related to frequent supervised vi physical activity trajectories as intact family predict less frequent non-supervised physical activity trajectories. Conclusion : First, involvement in supervised and non-supervised physical activity is relatively stable between 5 and 8 years old. Second, supervised and non-supervised physical activity appear to be two different processes that have their own set of predictors. Third, an ecological and multidimensional approach is required to capture the complexity of these two processes.

Traditionally design engineers have used the Factor of Safety method for ensuring that designs do not fail in the field. Access to advanced computational tools and resources have made this process obsolete and new methods to introduce higher levels of reliability in an engineering systems are currently being investigated. However, even though high computational resources are available the computational resources required by reliability analysis procedures leave much to be desired. Furthermore, the regression based surrogate modeling techniques fail when there is discontinuity in the design space, caused by failure mechanisms, when the design is required to perform under severe externalities. Hence, in this research we propose efficient Semi-Supervised Learning based surrogate modeling techniques that will enable accurate estimation of a system's response, even under discontinuity. These methods combine the available set of labeled dataset and unlabeled dataset and provide better models than using labeled data alone. Labeled data is expensive to obtain since the responses have to be evaluated whereas unlabeled data is available in plenty, during reliability estimation, since the PDF information of uncertain variables is assumed to be known. This superior performance is gained by combining the efficiency of Probabilistic Neural Networks (PNN) for classification and Expectation-Maximization (EM) algorithm for treating the unlabeled data as labeled data with hidden labels.

Introdução: Quedas representam risco extremamente incidente entre idosos, e sua recuperação produz altos custos. Algumas das causas mais comuns podem ser atenuadas por exercícios, se oferecidos de forma acessível. Objetivos: Comparar os efeitos de um treinamento físico multimodal quando realizado de forma supervisionada e semissupervisionada, sobre variáveis reconhecidamente relacionadas ao risco de quedas em idoso com preservada independência e histórico de quedas. Métodos: Setenta e seis idosos com histórico de quedas, acima de 70 anos, média de 79,06 anos (±4,55), foram avaliados sobre a saúde geral, histórico e risco de quedas, perigos domésticos, e foram submetidos aos seguintes testes: Timed up and Go (TUG), Walk Performance Test (WPT), Berg Balance Scale (BBS), avaliação isocinética do joelho e os seguintes testes em plataforma de equilibrio: Tandem Walk (TW), Sit to Stand (STS), Step up Over (SUO), Limits of Stability (LOS) e Modified Clinical Test of Sensory Integration on Balance (MCTSIB). Foram aleatoriamente alocados em 3 grupos: Supervisionado (S), orientado em todas as sessões, Semissupervisionado (SS), orientado quinzenalmente a executar exercícios em casa, e Controle (C), sem intervenção. O programa de exercícios multimodais foi executado em 3 sessões semanais de 50 minutos, por 4 meses. Participantes registraram quedas em calendário, e avaliações foram repetidas ao final do período. Resultados: Após intervenção o grupo S reduziu tempo do TUG (p<0,001) e no WPT (p<0,001) e aumentou a pontuação do BBS (p= 0,018), a Potência Média (p<0,001), o Pico de Torque/ Peso (p= 0,036) e a Média do Pico de Torque (p= 0,006) na flexão direita. Reduziu Tempo de Transferência no STS (p= 0,039), o Índice de Impacto na descida no SUO (p= 0.047), e a Oscilação no MCTSIB na 1ª (p= 0,037) e na 4ª (p= 0,032) condições avaliadas. No LOS, aumentou Velocidade (p<0,001), a Máxima Excursão (p<0,001) e o Controle de Direção (p= 0,004). O grupo SS reduziu o tempo no TUG (p= 0,001), aumentou o Índice de Fadiga na flexão do joelho direito (p= 0,043), aumentou Velocidade e reduziu Oscilação no TW (p= 0,008 e 0,020 respectivamente). No LOS, aumentou Velocidade (p= 0,023), a Máxima Excursão (p= 0,035) e o Controle de Direção (p= 0,006). O grupo C reduziu Velocidade no TW (p= 0,033) e aumentou o Índice de Fadiga na flexão direita (p= 0,017). O grupo S apresentou magnitude do efeito diferente na Potência Média da Flexão do Joelho direito (p= 0,002 para S versus SS, e p= 0,004 para S versus C). Os grupos S e C apresentaram diferença entre si na variação da Velocidade do LOS (p= 0,003). Os grupos S e SS obtiveram alterações diferentes do grupo C no TUG (p= 0,003 para C vs. S, e p= 0,021 para C vs. SS), e na Velocidade do TW (p= 0,007 para C vs. S, e p= 0,003 para C vs. SS). Conclusões: Numa população de idosos não institucionalizados, com independência preservada, baixa renda, pouca escolaridade, e com histórico de quedas, um treinamento físico multimodal, aplicado tanto de forma semissupervisionada, em casa, quanto de forma supervisionada, no centro de saúde, pode ser efetivo em melhorar variáveis previamente reconhecidas como sendo altamente relacionadas ao risco de quedas. Os resultados equivalentes entre os grupos S e SS impedem-nos de afirmar que a supervisão acrescente expressiva extensão a este benefício / Background: Falls are an extremely incidental healthcare risk among the geriatric populations and lead to high recuperative costs. Muscle weakness and balance impairment are among the most common causes and can be attenuated by exercises, if provided in an accessible way. Objectives: To compare the effects on variables related to falls risk, of a fully supervised center-based and a semi-supervised home-based multi-modal exercise program in elderly with preserved independence, and history of falls. Methods: Seventy six older adults with history of falls, over 70 years old, mean age of 79.06 years (±4.55) were assessed about general health, falls history and risk, home hazard and were submitted to the following tests: Timed up and Go (TUG), Walk Performance Test (WPT), Berg Balance Scale (BBS), Knee Isokinetic dynamometer test, and five tests on balance force plate: Tandem Walk (TW), Sit to Stand (STS), Step up Over (SUO), Limits of Stability (LOS) and modified Clinical Test of Sensory Integration on Balance (MCTSIB). Participants were randomized into three groups: supervised (S) that was instructed in all sessions, semi-supervised (SS) that received orientation every other week and performed the exercises at home, and control (C) that did not receive any exercise intervention. The multi-modal program consisted in three 50-minute sessions per week over four months. Participants recorded falls in a calendar and assessments were repeated at the end of the period Results: After intervention, S groups reduced time in TUG (p<0.001) and WPT (p<0.001), increased the BBS score (p= 0.018), the Average Power (p<0.001), the Peak Torque/Weight (p= 0.036), and the average Peak Torque (p= 0.006) of right knee flexion. It reduced Transfer Time in STS (p= 0.039), o Impact Index on SUO (p= 0.047), and End Sway on MCTSIB on 1st (p= 0.037) and 4th (p= 0.032) conditions assessed. On LOS, increased Movement Velocity (p<0.001), Maximum Excursion (p<0.001), and Directional Control (p= 0.004). The SS group reduced TUG (p= 0.001), increased Fatigue Work on right knee flexion (p= 0.043), increased Speed and reduced End Sway on TW (p= 0.008 e 0.020 respectively). On LOS, increased the velocity (p= 0,023), the Maximum Excursion (p= 0.035) and Directional Control (p= 0.006). The C group reduced TW speed (p= 0.033) and increased Fatigue Work of right knee flexion (p= 0.017). The S group showed different magnitude of effect in Average Power of right knee flexion (p= 0.002 for S vs. SS, and p= 0,004 for S vs. C). Groups S and C were different from each other on LOS Velocity (p= 0.003). Comparing to C, both trained groups, S and SS, had different magnitude of effect on TUG (p= 0.003 for C vs. S, and p= 0.021 for C vs. SS), and TW Speed (p= 0.007 for C vs. S, and p= 0.003 for C vs. SS). Conclusions: In a community-dwelling elderly population with preserved independence, low income and minimal education, with history of falls, a semi-supervised home-based and supervised center-based multi-modal exercise program, may be effective in improving variables previously recognized as highly related to falls risk. The similar results between trained groups prevent us to affirm that supervision adds expressive extent to the benefit