INTRODUÇÃO: O câncer de ovário é o mais letal de todos os cânceres ginecológicos e requer ser tratado por ginecologistas especializados em centros terciários para se obter melhor prognóstico. Este trabalho tem como objetivo analisar e comparar quatro estratégias diferentes para predizer a benignidade ou malignidade de tumores pélvicos supostamente de origem ovariana utilizando para este fim, marcadores tumorais CA 125 e HE4, índice de risco de malignidade (IRM) e algoritmo ROMA. MÉTODOS: Neste estudo prospectivo foram avaliadas 128 pacientes com diagnóstico de tumores pélvicos supostamente de origem ovariana atendidas na Divisão de Clínica Ginecológica do Hospital das Clinicas da Faculdade de Medicina da Universidade de São Paulo e Instituto do Câncer do Estado de São Paulo entre julho de 2008 e janeiro 2011. Foram calculadas a sensibilidade e a especificidade e construídas curvas ROC para comparar os quatro parâmetros (CA 125, HE4, ROMA e IRM) na eficácia de diferenciar tumores ovarianos. RESULTADOS: A sensibilidade obtida para CA 125, HE4, ROMA e IRM foi de, respectivamente, 70,4%, 79,7%, 74,1% e 63,0%. A especificidade para CA 125, HE4, ROMA e IRM foi de, respectivamente, 74,2%, 66,7%, 75,8% e 92,4%. Não houve diferença na comparação das áreas abaixo da curva ROC entre os quatro parâmetros. CONCLUSÕES: Nenhum dos quatro métodos estudados é o ideal na diferenciação de tumores ovarianos. Entre os quatro parâmetros analisados o HE4 foi o parâmetro com melhor sensibilidade na diferenciação de tumores ovarianos. A acurácia dos quatro métodos é equivalente e podem ser utilizados indistintamente para referenciar pacientes para serviços especializados no tratamento de câncer de ovário / BACKGROUND: Ovarian cancer is the most lethal of all gynecological cancers and requires to be treated by gynecologic oncologists in tertiary centers accustomed to treating this disease to achieve the best prognosis. This study aims to compare four different strategies to predict the benignity or malignancy of pelvic tumors presumably of ovarian origin using, for this purpose, tumor markers CA 125 and HE4, risk malignancy index (RMI) and algorithm ROMA. METHODS: This prospective study evaluated 128 patients supposedly with ovarian tumors treated at the Divisão de Clínica Ginecológica do Hospital das Clínicas da Faculdade de Medicina da Universidade de São Paulo and at Instituto do Câncer do Estado de São Paulo between July 2008 and January 2011. We calculated sensitivity, specificity and ROC curves to compare the four parameters (CA 125, HE4, ROMA and RMI) ability to differentiate the ovarian tumors. RESULTS: The sensitivity obtained for CA 125, HE4, ROMA and RMI was, respectively, 70.4%, 79.7%, 74.1% and 63.0%. The specificity obtained for CA 125, HE4, ROMA and RMI was, respectively, 74.2%, 66.7%, 75.8% and 92.4%. There was no difference the areas under the ROC curve among the four parameters. CONCLUSIONS: None of the four studied methods is best in the differentiation of ovarian tumors. Among the four parameters analyzed, HE4 was the parameter with highest sensitivity in the differentiation of ovarian tumors. The accuracy of the four methods is equivalent and can be used interchangeably to refer patients for specialized services in the treatment of ovarian cancer

INTRODUÇÃO: O câncer de ovário é o mais letal de todos os cânceres ginecológicos e requer ser tratado por ginecologistas especializados em centros terciários para se obter melhor prognóstico. Este trabalho tem como objetivo analisar e comparar quatro estratégias diferentes para predizer a benignidade ou malignidade de tumores pélvicos supostamente de origem ovariana utilizando para este fim, marcadores tumorais CA 125 e HE4, índice de risco de malignidade (IRM) e algoritmo ROMA. MÉTODOS: Neste estudo prospectivo foram avaliadas 128 pacientes com diagnóstico de tumores pélvicos supostamente de origem ovariana atendidas na Divisão de Clínica Ginecológica do Hospital das Clinicas da Faculdade de Medicina da Universidade de São Paulo e Instituto do Câncer do Estado de São Paulo entre julho de 2008 e janeiro 2011. Foram calculadas a sensibilidade e a especificidade e construídas curvas ROC para comparar os quatro parâmetros (CA 125, HE4, ROMA e IRM) na eficácia de diferenciar tumores ovarianos. RESULTADOS: A sensibilidade obtida para CA 125, HE4, ROMA e IRM foi de, respectivamente, 70,4%, 79,7%, 74,1% e 63,0%. A especificidade para CA 125, HE4, ROMA e IRM foi de, respectivamente, 74,2%, 66,7%, 75,8% e 92,4%. Não houve diferença na comparação das áreas abaixo da curva ROC entre os quatro parâmetros. CONCLUSÕES: Nenhum dos quatro métodos estudados é o ideal na diferenciação de tumores ovarianos. Entre os quatro parâmetros analisados o HE4 foi o parâmetro com melhor sensibilidade na diferenciação de tumores ovarianos. A acurácia dos quatro métodos é equivalente e podem ser utilizados indistintamente para referenciar pacientes para serviços especializados no tratamento de câncer de ovário / BACKGROUND: Ovarian cancer is the most lethal of all gynecological cancers and requires to be treated by gynecologic oncologists in tertiary centers accustomed to treating this disease to achieve the best prognosis. This study aims to compare four different strategies to predict the benignity or malignancy of pelvic tumors presumably of ovarian origin using, for this purpose, tumor markers CA 125 and HE4, risk malignancy index (RMI) and algorithm ROMA. METHODS: This prospective study evaluated 128 patients supposedly with ovarian tumors treated at the Divisão de Clínica Ginecológica do Hospital das Clínicas da Faculdade de Medicina da Universidade de São Paulo and at Instituto do Câncer do Estado de São Paulo between July 2008 and January 2011. We calculated sensitivity, specificity and ROC curves to compare the four parameters (CA 125, HE4, ROMA and RMI) ability to differentiate the ovarian tumors. RESULTS: The sensitivity obtained for CA 125, HE4, ROMA and RMI was, respectively, 70.4%, 79.7%, 74.1% and 63.0%. The specificity obtained for CA 125, HE4, ROMA and RMI was, respectively, 74.2%, 66.7%, 75.8% and 92.4%. There was no difference the areas under the ROC curve among the four parameters. CONCLUSIONS: None of the four studied methods is best in the differentiation of ovarian tumors. Among the four parameters analyzed, HE4 was the parameter with highest sensitivity in the differentiation of ovarian tumors. The accuracy of the four methods is equivalent and can be used interchangeably to refer patients for specialized services in the treatment of ovarian cancer

Predictive power of classification models can be evaluated by various measures. The most popular measures in data mining (DM) are Gini coefficient, Kolmogorov-Smirnov statistic and lift. These measures are each based on a completely different way of calculation. If an analyst is used to one of these measures it can be difficult for him to asses the predictive power of a model evaluated by another measure. The aim of this thesis is to develop a method how to convert one performance measure into another. Even though this thesis focuses mainly on the above-mentioned measures, it deals also with other measures like sensitivity, specificity, total accuracy and area under ROC curve. During development of DM models you may need to work with a sample that is stratified by values of the target variable Y instead of working with the whole population containing millions of observations. If you evaluate a model developed on a stratified data you may need to convert these measures to the whole population. This thesis describes a way, how to carry out this conversion. A software application (CPM) enabling all these conversions makes part of this thesis. With this application you can not only convert one performance measure to another, but you can also convert measures calculated on a stratified sample to the whole population. Besides the above mentioned performance measures (sensitivity, specificity, total accuracy, Gini coefficient, Kolmogorov-Smirnov statistic), CPM will also generate confusion matrix and performance charts (lift chart, gains chart, ROC chart and KS chart). This thesis comprises the user manual to this application as well as the web address where the application can be downloaded. The theory described in this thesis was verified on the real data.

INTRODUÇÃO: O Hipotireoidismo congênito (HC) é uma das endocrinopatias mais frequentes em pediatria e pode causar retardo mental e do crescimento, se não for tratado precocemente. A determinação do nível do hormônio estimulador da tireoide em sangue total após o nascimento (TSHneo) constitui uma estratégia efetiva para o rastreamento de HC, embora não exista consenso em relação aos níveis considerados seguros para essa detecção. Muitos serviços utilizam os valores de corte do TSH neonatal de 10,0 e 15,0 ?UI/mL, por ensaios imunofluorimétricos. OBJETIVO: Analisar a capacidade de detecção dos casos de hipotireoidismo congênito por diferentes níveis de corte do TSH neonatal e os efeitos destes sobre o sistema de triagem neonatal para essa doença, em nascidos vivos avaliados pelo Programa de Triagem Neonatal (PTN) da rede pública do Estado de Mato Grosso (MT), de 01 de janeiro de 2010 a 31 de dezembro de 2012. MÉTODOS: Estudo de coorte, de corte transversal, com coleta retrospectiva de dados obtidos a partir do banco de dados do Serviço de Referência em Triagem Neonatal do Estado de Mato Grosso, de nascidos vivos no período 01/01/2010 a 31/12/2012 e avaliados pelo PTN-MT. Estes foram divididos em dois grupos: I. Controle: Crianças com exame de triagem neonatal normal; II. Estudo: Crianças com HC. Análise estatística incluiu o uso do teste qui-quadrado ou exato de Fischer para análise das características dos recém-nascidos entre os grupos e o teste t de Student ou não paramétrico de Mann-Whitney para análise dos níveis de TSH em sangue total de ambos os grupos e, avaliação das concentrações de TSH e T4 livre no soro, em crianças com HC. Construiu-se uma curva ROC (Receiver Operating Characteristic), para a avaliação dos pontos de corte do TSHneo. O nível de significância foi p<0,05. RESULTADOS: Entre as 111.705 crianças triadas pelo Programa, 50 tiveram o diagnóstico de HC, sob o ponto de corte do TSHneo de 5,0 ?UI/mL. A prevalência da doença foi de 1:2.234 nascidos vivos. A cobertura do Programa estadual foi de 73,9%. Para o Grupo II, os níveis do TSHneo foram superiores a 20,0 ?UI/mL em 61,4% das crianças e, os níveis de TSH no soro excederam este valor em 83,7%. A curva ROC identificou o ponto de corte do TSHneo de 5,03 ?UI/mL, como o correspondente à sensibilidade de 100% e a maior especificidade associada (93,7%). A área observada sob a curva foi de 0,9898 (p<0,0001). CONCLUSÕES: Observou-se uma cobertura inadequada do PTN-MT. O ponto de corte do TSH neonatal de 5,0 ?UI/mL, adotado pelo PTN-MT, foi confirmado pela curva ROC como o mais seguro para detectar HC e determinou a elevada prevalência da doença no Estado de Mato Grosso / INTRODUCTION: Congenital hypothyroidism (CH) is a very common pediatric endocrine disorder and can cause mental and growth retardation without early treatment. Measuring the total blood thyroid-stimulating hormone level after birth (TSHneo) is an effective screening strategy for CH, although there is not yet a consensus on the appropriate diagnostic levels. Many services use the neonatal TSH cut-off points of 10.0 and 15.0 uIU/mL per imunofluorimetric assay. OBJECTIVE: The aim of the present study was to analyze the ability of various TSHneo cutoff values to detect CH and their effects on the Newborn Screening Program (NSP) of the State of Mato Grosso (MT) from January 1, 2010, to december 31, 2012. METHODS: Cohort study, cross-sectional, based on retrospective data collection obtained from the database of the Reference Service for Neonatal Screening of the State of Mato Grosso, for all live births from January 1, 2010, to December 31, 2012, reviewed by NSP-MT. The infants were divided into two groups: I-Control: infants with normal newborn screening tests and II-Study: infants with CH. Statistical analysis included the chi-square or Fisher\'s exact test to compare the characteristics of the newborns from both groups and Student\'s t-test or the non-parametric Mann-Whitney test to analyse the total blood TSH level from both groups of infants and evaluate the serum TSH and free thyroxine (T4) concentrations in infants with CH. A Receiver Operating Characteristic (ROC) curve was constructed to assess the TSHneo cutoff values. The significance level was p < 0.05. RESULTS: Using a TSHneo cutoff value 5.0 uIU/mL, 50 out of 111,705 screened infants were diagnosed with CH. The prevalence of CH was 1:2,234 live births. The state program coverage was 73.9%. For Group II, the TSHneo levels were higher than 20.0 uIU/mL in 61.4% of infants, and the serum TSH levels exceeded that level in 83.7%. The ROC curve showed that a TSHneo cutoff value of 5.03 uIU/mL had 100% sensitivity and the greatest associated specificity (93.7%). The area under the curve was 0.9898 (p < 0.0001). CONCLUSIONS: An inadequate coverage of the NSP-MT was observed. The ROC curve confirmed that the TSHneo cutoff value of 5.0 ?IU/mL adopted by the NSP-MT was the safest for detecting CH and determined the high prevalence of disease that was found in the State of Mato Grosso

<p>Akutni pankreatitis predstavlja zbirni pojam dinamičkih, lokalnih i sistemskih patofiziolo&scaron;kih procesa nastalih iznenadnim prodorom aktivnih litičkih pankreasnih enzima u žlezdani parenhim. Cilj istraživanja je da se Utvrditi senzitivnost difuzione sekvence magnetne rezonance (DWI) radi utvrđivanja morfolo&scaron;kih promena parenhima kod akutnog pankreatitisa. Poređenje difuzione mape i difuzionog koeficijenta kod pacijenata sa akutnim pankreatitisom i kod pacijenata sa morfolo&scaron;ki urednim parenhimom pankreasa na magnetnoj rezonanci. Utvrditi da li postoje statistički značajne razlike difuzionog koeficijenta kod pacijenata sa akutnim pankreatitisom u odnosu na pol. Utvrditi da li postoje statistički značajne razlike difuzionog koeficijenta kod pacijenata sa akutnim pankreatitisom u odnosu na godine. Odrediti prelomnu tačku difuzionog koeficijenta kod pacijenata sa akutnim pankreatitisom. Studija je bila prospektivnog karaktera i obuhvatilo je 30 ispitanika sa morfolo&scaron;ki urednim parenhimom pankreasa i 30 sa dijagnozom akutnog pankreatitisa unutar 72 sata od početka simptoma. Svi pacijenti su pregledani magnetnom rezonancom u Centru za radiologiju, Kliničkog Centra Vojvodine. Rezultati ukazuju da postoje razlike difuzionog koeficijenta kod pacijenata sa akutnim pankreatitisom i kontrolne grupe. Takođe smo dokazali da difuzioni koeficijent zavisi od pola i starosti i utvrdili smo prelomnu tačku difuzije za rano dijagnostikovanje akutnog pankreatitisa.</p> / <p>Acute pancreatitis is defined as cumulative term of dynamic local and general pathophysiological processes caused by sudden penetration of active lithic pancreatic enzymes in the glandular parenchyma. Goal of this research is to note the changes (sensitivity) in values of diffusion weighted images (DWI) in acute pancreatitis and to determine morphological changes in glandular parenchyma of pancreas. Comparation of DWI between patients with acute pancreatitis and patients with normal pancreatic parenchyma based on magnetic resonance (MRI). We also want to determine whether there were statistically significant differences of DWI in patients with acute pancreatitis in relation to sex and age. One of our goals also was to determine breakpoint of DWI as a sure sign of acute pancreatitis. This was prospective study and included 30 patients with morphologically healthy parenchyma of the pancreas (control group) and 30 with the diagnosis of acute pancreatitis &ndash; in first 72 hours of the onset of symptoms. All patients were examined on MRI in department of Radiology of Clinical Center of Vojvodina. Our results indicate that was a big difference of DWI between patients with acute pancreatitis and control group. We prove that DWI depends on the sex and age. 1,77x10-6mm/s2 was breakpoint which indicates acute pancreatitis.</p>

Οι αλγόριθμοι μηχανικής μάθησης είναι επιθυμητό να είναι σε θέση να γενικεύσουν για οποιασδήποτε κλάση με ίδια ακρίβεια. Δηλαδή σε ένα πρόβλημα δύο κλάσεων - θετικών και αρνητικών περιπτώσεων - ο αλγόριθμος να προβλέπει με την ίδια ακρίβεια και τα θετικά και τα αρνητικά παραδείγματα. Αυτό είναι φυσικά η ιδανική κατάσταση. Σε πολλές εφαρμογές οι αλγόριθμοι καλούνται να μάθουν από ένα σύνολο στοιχείων, το οποίο περιέχει πολύ περισσότερα παραδείγματα από τη μια κλάση σε σχέση με την άλλη. Εν γένει, οι επαγωγικοί αλγόριθμοι είναι σχεδιασμένοι να ελαχιστοποιούν τα σφάλματα. Ως συνέπεια οι κλάσεις που περιέχουν λίγες περιπτώσεις μπορούν να αγνοηθούν κατά ένα μεγάλο μέρος επειδή το κόστος λανθασμένης ταξινόμησης της υπερ-αντιπροσωπευόμενης κλάσης ξεπερνά το κόστος λανθασμένης ταξινόμησης της μικρότερη κλάση. Το πρόβλημα των ανομοιογενών συνόλων δεδομένων εμφανίζεται και σε πολλές πραγματικές εφαρμογές όπως στην ιατρική διάγνωση, στη ρομποτική, στις διαδικασίες βιομηχανικής παραγωγής, στην ανίχνευση λαθών δικτύων επικοινωνίας, στην αυτοματοποιημένη δοκιμή του ηλεκτρονικού εξοπλισμού, και σε πολλές άλλες περιοχές. Η παρούσα διπλωματική εργασία με τίτλο ‘Μηχανική Μάθηση με Ανομοιογενή Δεδομένα’ (Machine Learning with Imbalanced Data) αναφέρεται στην επίλυση του προβλήματος αποδοτικής χρήσης αλγορίθμων μηχανικής μάθησης σε ανομοιογενή/ανισοκατανεμημένα δεδομένα. Η διπλωματική περιλαμβάνει μία γενική περιγραφή των βασικών αλγορίθμων μηχανικής μάθησης και των μεθόδων αντιμετώπισης του προβλήματος ανομοιογενών δεδομένων. Παρουσιάζεται πλήθος αλγοριθμικών τεχνικών διαχείρισης ανομοιογενών δεδομένων, όπως οι αλγόριθμοι AdaCost, Cost Senistive Boosting, Metacost και άλλοι. Παρατίθενται οι μετρικές αξιολόγησης των μεθόδων Μηχανικής Μάθησης σε ανομοιογενή δεδομένα, όπως οι καμπύλες διαχείρισης λειτουργικών χαρακτηριστικών (ROC curves), καμπύλες ακρίβειας (PR curves) και καμπύλες κόστους. Στο τελευταίο μέρος της εργασίας προτείνεται ένας υβριδικός αλγόριθμος που συνδυάζει τις τεχνικές OverBagging και Rotation Forest. Συγκρίνεται ο προτεινόμενος αλγόριθμος σε ένα σύνολο ανομοιογενών δεδομένων με άλλους αλγόριθμους και παρουσιάζονται τα αντίστοιχα πειραματικά αποτελέσματα που δείχνουν την καλύτερη απόδοση του προτεινόμενου αλγόριθμου. Τελικά διατυπώνονται τα συμπεράσματα της εργασίας και δίνονται χρήσιμες ερευνητικές κατευθύνσεις. / Machine Learning (ML) algorithms can generalize for every class with the same accuracy. In a problem of two classes, positive (true) and negative (false) cases-the algorithm can predict with the same accuracy the positive and negative examples that is the ideal case. In many applications ML algorithms are used in order to learn from data sets that include more examples from the one class in relationship with another class. In general inductive algorithms are designed in such a way that they can minimize the occurred errors. As a conclusion the classes that contain some cases can be ignored in a large percentage since the cost of the false classification of the super-represented class is greater than the cost of false classification of lower class. The problem of imbalanced data sets is occurred in many ‘real’ applications, such as medical diagnosis, robotics, industrial development processes, communication networks error detection, automated testing of electronic equipment and in other related areas. This dissertation entitled ‘Machine Learning with Imbalanced Data’ is referred to the solution of the problem of efficient use of ML algorithms with imbalanced data sets. The thesis includes a general description of basic ML algorithms and related methods for solving imbalanced data sets. A number of algorithmic techniques for handling imbalanced data sets is presented, such as Adacost, Cost Sensitive Boosting, Metacost and other algorithms. The evaluation metrics of ML methods for imbalanced datasets are presented, including the ROC (Receiver Operating Characteristic) curves, the PR (Precision and Recall) curves and cost curves. A new hybrid ML algorithm combining the OverBagging and Rotation Forest algorithms is introduced and the proposed algorithmic procedure is compared with other related algorithms by using the WEKA operational environment. Experimental results demonstrate the performance superiority of the proposed algorithm. Finally, the conclusions of this research work are presented and several future research directions are given.

Objectif : Déterminer la fiabilité et la précision d’un prototype d’appareil non invasif de mesure de glucose dans le tissu interstitiel, le PGS (Photonic Glucose Sensor), en utilisant des clamps glycémiques multi-étagés. Méthodes : Le PGS a été évalué chez 13 sujets avec diabète de type 1. Deux PGS étaient testés par sujet, un sur chacun des triceps, pour évaluer la sensibilité, la spécificité, la reproductibilité et la précision comparativement à la technique de référence (le Beckman®). Chaque sujet était soumis à un clamp de glucose multi-étagé de 8 heures aux concentrations de 3, 5, 8 et 12 mmol/L, de 2 heures chacun. Résultats : La corrélation entre le PGS et le Beckman® était de 0,70. Pour la détection des hypoglycémies, la sensibilité était de 63,4%, la spécificité de 91,6%, la valeur prédictive positive (VPP) 71,8% et la valeur prédictive négative (VPN) 88,2%. Pour la détection de l’hyperglycémie, la sensibilité était de 64,7% et la spécificité de 92%, la VPP 70,8% et la VPN : 89,7%. La courbe ROC (Receiver Operating Characteristics) démontrait une précision de 0,86 pour l’hypoglycémie et de 0,87 pour l’hyperglycémie. La reproductibilité selon la « Clark Error Grid » était de 88% (A+B). Conclusion : La performance du PGS était comparable, sinon meilleure que les autres appareils sur le marché(Freestyle® Navigator, Medtronic Guardian® RT, Dexcom® STS-7) avec l’avantage qu’il n’y a pas d’aiguille. Il s’agit donc d’un appareil avec beaucoup de potentiel comme outil pour faciliter le monitoring au cours du traitement intensif du diabète. Mot clés : Diabète, diabète de type 1, PGS (Photonic Glucose Sensor), mesure continue de glucose, courbe ROC, « Clark Error Grid». / Objective: To determine the reliability and precision of a prototype of a non-invasive device for continuous measurement of interstitial glucose, the PGS (Photonic Glucose Sensor), using multi-level glycaemic clamp. Methods: The PGS was evaluated in 13 subjects with type 1 diabetes. Two PGS were tested with each subject, one on each triceps, to evaluate the sensitivity, specificity, reproducibility and accuracy compared to the reference technique, the glucose analyzer Beckman®. Each subject was submitted to a multi-level 8 hour glucose clamp at 3, 5, 8 and 12 mmol / L, 2 hours each. Results: The correlation between the PGS and the Beckman® was 0.70. For the detection of hypoglycaemia, the sensitivity was 63.4%, the specificity 91.6%, the positive predictive value (PPV) 71.8% and the negative predictive value (NPV) 88.2%. For the detection of hyperglycaemia, the sensitivity was 64.7% the specificity 92%, the PPV 70.8% and the NPV: 89.7%. The ROC (Receiver Operating Characteristics) curve showed an accuracy of 0.86 and 0.87 for hypoglycaemia and hyperglycaemia respectively. Reproducibility according to the Clark Error Grid was 88% in the A and B zone. Conclusion: The performance of the PGS was comparable or better than other continuous glucose monitoring devices on the market (Freestyle® Navigator, Medtronic Guardian® RT, Dexcom® STS-7) with the advantage that it has no needle. It is therefore an interesting device and hopefully, which could facilitate the monitoring in the intensive treatment of diabetes. Key words: Diabetes, type 1 diabetes, PGS (Photonic Glucose Sensor), ROC curve, Clark Error Grid, continuous glucose monitoring, CGMS.

Huit instruments d’évaluation du risque ont été appliqués sur 580 délinquants sexuels. Il s’agit du VRAG, du SORAG, du RRASOR, de la Statique-99, de la Statique-2002, du RM-2000, du MnSORT-R et du SVR-20. De plus, les sujets ont été cotés sur la PCL-R, qui vise la mesure de la psychopathie, mais qui a fait ses preuves en matière de prédiction de la récidive (Gendreau, Little, et Goggin, 1996). En vue de mesurer l’efficacité de ces instruments et de la PCL-R, une période de suivi de 25 ans a été observée. Aussi, une division de l’échantillon a été faite par rapport à l’âge au moment de la libération, afin de mesurer les différences entre les délinquants âgés de 34 ans et moins et ceux de 35 ans et plus. Le présent travail vise à répondre à trois objectifs de recherche, soit 1) Décrire l’évolution du risque en fonction de l’âge, 2) Étudier le lien entre l’âge, le type de délinquant et la récidive et 3) Comparer l’efficacité de neuf instruments structurés à prédire quatre types de récidive en fonction de l’âge. Les résultats de l’étude suggèrent que l’âge influence le niveau de risque représenté par les délinquants. Par ailleurs, les analyses des différents types de récidive indiquent que le type de victime privilégié par les délinquants influence également ce niveau de risque. Les implications théoriques et pratiques seront discutées. / Eight evaluation techniques demonstrating high risk sexual offenders has been taken upon 580 individual sexual offenders. They are among the VRAG, the SORAG, the RRASOR, the Static-99, the Static-2002, the RM-2000, the MnSORT-R ans the SVR-20. Also, the subjects have been quoted according to the PCL-R, which focuses on their mental health, however supporting the quotes by prediction and relaps (Gendreau, Little, et Goggin, 1996). With the ongoing measuring of the suitability of these instruments as well as the PCL-R, it will take a period of 25 years for the observance. As well, a group of subjects have been studied from the age they were let out of prison, to come up with conclusions differentiating the offenders aging 34 and less with the offenders aging 35 and older. The present work on this subject matter hopes to focus on three research objectives: 1) Describe the evolution of risk according to the age, 2) To study the common point between age, they type of offender, and their relaps, and 3) To compare the suitability of 9 instruments the predict 4 types of relaps according to their age. The study results suggest that age affects the level of risk posed by offenders. Furthermore, analyses of different types of recidivism indicate that the preferred type of victim offender also influences the level of risk. The theoretical and practical implications are discussed.

Nous proposons dans cette thèse la méthode STREE de construction d'arbres de décision avec des données symboliques. Ce type de données permet de caractériser des individus de niveau supérieur qui peuvent être des classes ou catégories d’individus ou des concepts au sens des treillis de Galois. Les valeurs des variables, appelées variables symboliques, peuvent être des ensembles, des intervalles ou des histogrammes. Le critère de partitionnement récursif est une combinaison d'un critère par rapport aux variables explicatives et d'un critère par rapport à la variable à expliquer. Le premier critère est la variation de la variance des variables explicatives. Quand il est appliqué seul, STREE correspond à une méthode descendante de classification non supervisée. Le second critère permet de construire un arbre de décision. Il s'agit de la variation de l'indice de Gini si la variable à expliquer est nominale et de la variation de la variance si la variable à expliquer est continue ou bien est une variable symbolique. Les données classiques sont un cas particulier de données symboliques sur lesquelles STREE peut aussi obtenir de bons résultats. Il en ressort de bonnes performances sur plusieurs jeux de données UCI par rapport à des méthodes classiques de Data Mining telles que CART, C4.5, Naive Bayes, KNN, MLP et SVM. STREE permet également la construction d'ensembles d'arbres de décision symboliques soit par bagging soit par boosting. L'utilisation de tels ensembles a pour but de pallier les insuffisances liées aux arbres de décisions eux-mêmes et d'obtenir une décision finale qui est en principe plus fiable que celle obtenue à partir d'un arbre unique. / In this thesis, we propose the STREE methodology for the construction of decision trees with symbolic data. This data type allows us to characterize individuals of higher levels which may be classes or categories of individuals or concepts within the meaning of the Galois lattice. The values of the variables, called symbolic variables, may be sets, intervals or histograms. The criterion of recursive partitioning is a combination of a criterion related to the explanatory variables and a criterion related to the dependant variable. The first criterion is the variation of the variance of the explanatory variables. When it is applied alone, STREE acts as a top-down clustering methodology. The second criterion enables us to build a decision tree. This criteron is expressed as the variation of the Gini index if the dependant variable is nominal, and as the variation of the variance if thedependant variable is continuous or is a symbolic variable. Conventional data are a special case of symbolic data on which STREE can also get good results. It has performed well on multiple sets of UCI data compared to conventional methodologies of Data Mining such as CART, C4.5, Naive Bayes, KNN, MLP and SVM. The STREE methodology also allows for the construction of ensembles of symbolic decision trees either by bagging or by boosting. The use of such ensembles is designed to overcome shortcomings related to the decisions trees themselves and to obtain a finaldecision that is in principle more reliable than that obtained from a single tree.

This diploma thesis is focused on the evaluation of the efficiency of selected bankruptcy models in the Czech Republic. In the theoretical part the basic terminology and methodology of bankruptcy models creation are introduced. In addition are mentioned, model constraints, an overview of the indicators used, and information about model accuracy. This part also presents analyzed models and methods of assessing the reliability of bankruptcy models. In the practical part, the reliability of selected bankruptcy models is evaluated and a new bankruptcy model is built.