The United States public schools promote inclusion and educational equity among diverse student populations. Considerable and growing numbers of students with categorical disabilities and Limited English Proficiency (LEP) are enrolled in regular classrooms. The systemic barriers in learning that they have could impact teacher perceptions and decisions about teaching practices as well as the teaching profession. These students have challenged K-12 science, technology, engineering, and mathematics (STEM) teachers to provide high-quality, accommodative service and equitable educational opportunities in an increasingly STEM-infused society. Professional development associated with teaching students with disabilities and LEP is critical to inform in-service STEM teachers with these students' learning needs and promote student success. Effective preparation and support help maintain teacher satisfaction and retention within the teaching profession. However, the levels and perceptions of STEM teacher participation in such professional development, and whether the service load and professional development regarding the concerned groups of students associated with teacher satisfaction and retention remain unclear. This dissertation addresses these issues through two research studies using secondary analysis of the 2011-2012 School and Staffing Survey Teacher Questionnaire (SASS TQ) national dataset. The first study focused on K-12 STEM educator participation and perceived utility regarding their professional development experience concerning students with disabilities and LEP. Quantitative analysis revealed an overall lower level of participation and perceived utility of such professional development for STEM educators compared to all other educators. The second study examined teacher satisfaction and intent to remain in teaching, as well as their relationships to teacher service load and professional development specific to students with disabilities and LEP. Results indicated that K-12 STEM educators were less likely to feel satisfied or intent on remaining in teaching, compared to the remainder of the teaching population. Logistic regressions showed that service load of students with LEP predicted teacher satisfaction and participation in professional development concerning students with disabilities associated with teacher intent to remain in STEM education, especially for science educators. These findings collectively suggested the necessity and demands of sufficient and useful professional development offerings regarding the two concerned groups of students in inclusive STEM education settings. / Ph. D.

Heterogeneous data with different characteristics are ubiquitous in the modern digital world. For example, the observations collected from a process may change on its mean or variance. In numerous applications, data are often of mixed types including both discrete and continuous variables. Heterogeneity also commonly arises in data when underlying models vary across different segments. Besides, the underlying pattern of data may change in different dimensions, such as in time and space. The diversity of heterogeneous data structures makes statistical modeling and analysis challenging. Detection of change-points in heterogeneous data has attracted great attention from a variety of application areas, such as quality control in manufacturing, protest event detection in social science, purchase likelihood prediction in business analytics, and organ state change in the biomedical engineering. However, due to the extraordinary diversity of the heterogeneous data structures and complexity of the underlying dynamic patterns, the change-detection and analysis of such data is quite challenging. This dissertation aims to develop novel statistical modeling methodologies to analyze four types of heterogeneous data and to find change-points efficiently. The proposed approaches have been applied to solve real-world problems and can be potentially applied to a broad range of areas. / Ph. D.

Grape root borer, Vitacea polistiformis (Harris), is an oligophagous pest of grapevines in the eastern USA. Neonates must burrow into the soil to find grape roots. In Virginia, larvae feed on roots for ~2 years, then pupate just beneath the soil surface. Emerging adults leave an empty pupal exuviae at the soil surface around the vine base. There was no relationship between weekly captures in pheromone traps and pupal exuviae counts, indicating that exuviae sampling is most appropriate to assess infestations. Exuviae sampling in Virginia vineyards revealed infestations that ranged from light to very heavy. Eighteen biotic and abiotic variables were measured and used in analyses that assessed their relative contributions to differences in exuviae density. Water holding capacity and clay/sand ratio were most strongly associated with pupal exuviae density; these variables were used to develop a model for predicting the extent of infestation of individual vineyards. The spatial distribution of pupal exuviae was characterized using non-spatial and geospatial techniques. Although the non-spatial method (Taylor's Power Law) indicated that exuviae showed an aggregated distribution in all blocks, spatial methods (variograms, SADIE) revealed aggregated distributions only in blocks with ≥ 0.5 pupal exuviae per vine. Independent pupal exuviae samples for population assessment in vineyards can be achieved using sampling points separated by >8.8 m. Combined results from geospatial analyses and the temporal distribution of pupal exuviae within years enabled the development of a practical and quantitative sampling protocol. Bioassays used to measure the behavioral response of larvae to host stimuli revealed that neonates were attracted to grape root volatiles. In soil column bioassays, larvae moved vertically and horizontally over distances of up to 120 cm and apparently perceived the presence of grape roots from a distance of 5 cm in soil. Results are discussed in relation to their potential implications for monitoring and managing grape root borer. / Ph. D.

In this project, we explored the usage of machine learning in classifying portable electronic devices. The primary objective was to identify devices such as laptops, smartphones, and tablets, based on their physical and technical specification. These specifications, sourced from the Pricerunner price comparison website, contain height, Wi-Fi standard, and screen resolution. We aggregated this information into a dataset and split it into a training set and a testing set. To achieve the classification of devices, we trained four popular machine learning models: Random Forest (RF), Logistic Regression (LR), k-Nearest Neighbor (kNN), and Fully Connected Network (FCN). We then compared the performance of these models. The evaluation metrics used to compare performance included precision, recall, F1-score, accuracy, and training time. The RF model achieved the highest overall accuracy of 95.4% on the original dataset. The FCN, applied to a dataset processed with standardization followed by Principal Component Analysis (PCA), reached an accuracy of 92.7%, the best within this specific subset. LR excelled in a few class-specific metrics, while kNN performed notably well relative to its training time. The RF model was the clear winner on the original dataset, while the kNN model was a strong contender on the PCA-processed dataset due to its significantly faster training time compared to the FCN. In conclusion, the RF was the best-performing model on the original dataset, the FCN showed impressive results on the standardized and PCA-processed dataset, and the kNN model, with its highest macro precision and rapid training time, also demonstrated competitive performance.

This thesis is a case study conducted at Magna Electronics to explore the use of machinelearning techniques in improving the predictive quality of electronic control unit (ECU)within the automotive manufacturing. This thesis aims to apply interpretable machinelearning methods to predict potential future ECU failures early. With the interpretablemachine learning the goal is to identify predictive variables that lead to ECU failure andwhich can be used as support for decision making.Logistic Regression and Random Forest were chosen as the machine learning methods,which have been used in research of predictive quality and have different levels of interpretability.TreeSHAP was used on the Random Forest as the post-hoc method to furtherunderstand the results. The models’ performances were quantitatively evaluatedthrough metrics such as accuracy and area under precision-recall curve. Subsequently, thebest-performing models were further analyzed using confusion matrices, precision-recallcurves, and horizontal bar charts to assess the impact of predictive variables.The results of this thesis indicated that while Random Forest outperformed Logistic Regression,both models demonstrated limited capability in accurately predicting faulty ECUs,due to the low AUCPR scores. The precision-recall curves suggested performance near randomguess, highlighting the possible variability in parameter impact.This study has also identified significant challenges, such as data imbalance and mislabeling,which may have had a negative effect on the results. Given these issues, the thesisadvises caution in using these results for decision-making. Although, findings of this thesisunderscore the need for a cautious approach to interpreting model outputs, suggestingthat real-world application may require to use different models based on the specific goalsand context of the analysis.

In recent years, there has been an increase in the issue of bullying, particularly in academic settings. This degree project examines the use of supervised machine learning techniques to identify bullying in text data from school surveys provided by the Friends Foundation. It evaluates various traditional algorithms such as Logistic Regression, Naive Bayes, SVM, Convolutional neural networks (CNN), alongside a Retrieval-Augmented Generation (RAG) model using Llama 3, with a primary goal of achieving high recall on the texts consisting of bullying while also considering precision, which is reflected in the use of the F3-score. The SVM model emerged as the most effective among the traditional methods, achieving the highest F3-score of 0.83. Although the RAG model showed promising recall, it suffered from very low precision, resulting in a slightly lower F3-score of 0.79. The study also addresses challenges such as the small and imbalanced dataset as well as emphasizes the importance of retaining stop words to maintain context in the text data. The findings highlight the potential of advanced machine learning models to significantly assist in bullying detection with adequate resources and further refinement.

Le cancer de la prostate est l’un des cancers les plus fréquents chez les hommes. Le rôle de l’obésité dans son étiologie revêt un intérêt grandissant. Les associations observées sont souvent contradictoires selon le type d’obésité (générale ou abdominale), la durée ou le moment de l’exposition et l’agressivité du cancer. L’obésité abdominale figure parmi les pistes de recherche particulièrement prometteuses. L’objectif général de cette thèse était d’examiner la relation entre l’obésité et le cancer de la prostate, tant au niveau individuel que familial. Nous avons utilisé les données de PROtEuS, une étude cas-témoins populationnelle conduite en 2005-2012 à Montréal. Cette étude comprend un large éventail d’informations anthropométriques et liées aux habitudes de vie, recueillies chez 1931 cas et 1994 témoins de la population générale, ainsi que des informations anthropométriques relatives aux membres de leurs familles respectives. Les deux premiers objectifs spécifiques de la thèse visaient à explorer la corrélation entre différents indicateurs d’obésité et à développer une approche alternative à la mesure directe pour décrire l’obésité abdominale au moyen de modèles prédictifs. Les résultats suggèrent que les silhouettes de Stunkard et Sorensen sont étroitement liées à l’indice de masse corporelle (IMC) et au poids rapporté, tant au moment de l’entrevue que dans le passé. Nous avons montré qu’il était possible de prédire l’obésité abdominale relativement bien (R2=0.64), plus particulièrement la circonférence de la taille, à partir de l’IMC, la silhouette et la taille de pantalon. Les objectifs spécifiques 3 et 4 visaient à examiner l’association entre différents indicateurs d’obésité individuelle à plusieurs âges, ainsi que les trajectoires adultes d'obésité générale et abdominale, et le risque de cancer de la prostate. Nos résultats suggèrent un risque réduit de cancer de la prostate chez les personnes en surpoids ou obèses (rapport de cotes (RC) 0,71; intervalle de confiance à 95% (IC 95%) 0,59 - 0,85). À l’opposé, l’obésité abdominale récente, basée sur plusieurs indicateurs, était associée à un risque accru de cancer de haut grade (RC 1,33; IC 95% 1,03 - 1,71 pour une circonférence de la taille ≥ 102 cm). Les objectifs spécifiques 5 et 6 visaient à évaluer le risque de récurrence familiale d’obésité en fonction du nombre de membres de la famille atteints d’un cancer de la prostate et d’évaluer la co-agrégation familiale de l’obésité et du cancer de la prostate, indépendamment des agrégations familiales de l’obésité et du cancer de la prostate elles-mêmes. Le risque de récurrence familiale d’obésité était plus élevé lorsque deux cas ou plus de cancer de la prostate étaient observés dans la famille qu’en l’absence de cancer et ce, peu importe le nombre de cas d’obésité dans la famille. Pour les familles avec 3 parents présentant une obésité, le risque de récurrence d’obésité était de 0,35 (IC 95% 0,32 – 0,37) pour ceux n’ayant aucun cas de cancer dans la famille et de 0,38 (IC 95% 0,33 – 0,43) pour ceux avec au moins 2 cas de cancer dans la famille. La différence entre les risques de récurrence était encore plus marquée lorsque les familles issues du témoin index étaient comparées à celles dont le cas index avaient une tumeur de grade élevé au diagnostic. Pour les familles avec 3 parents présentant une obésité, le risque de récurrence d’obésité était de 0,35 (IC 95% 0,32 – 0,37) pour ceux dont le participant index était un témoin et de 0,41 (IC 95% 0,36 – 0,45) pour ceux dont le cas index avait une tumeur de grade élevé au moment du diagnostic. Nous n’avons pas observé de co-agrégation familiale entre obésité et cancer de la prostate dans son ensemble. Toutefois, cette co-agrégation était présente pour les cancers apparaissant avant l’âge de 55 ans (RC 1,35; IC 95% 1,11 - 1,65). Les résultats de cette thèse renforcent l’hypothèse d’un rôle important de l’obésité dans le développement du cancer de la prostate. Observation fort novatrice, l’obésité semble plus fréquente dans les familles avec plusieurs diagnostics précoces ou tumeurs agressives. La confirmation du rôle de l’obésité, facteur modifiable, dans l’étiologie de ce cancer très répandu aurait des retombées substantielles sur la santé publique. / Prostate cancer is among the most frequently diagnosed solid tumor among men. The role of obesity in its etiology is of mounting interest. The associations observed have been sometimes contradictory, varying according to the type of obesity (general or abdominal), the duration or the moment of exposure, and cancer aggressiveness. Abdominal obesity represents a promising research avenue. The general objective of this thesis was to examine the relationship between obesity and prostate cancer, both at the individual and familial levels. The main data source used was PROtEuS, a population-based case-control study conducted in 2005-2012 in Montreal. It collected a wide range of anthropometric and lifestyle information from 1,931 cases and 1,994 population controls, as well as information on obesity among their respective family members. The first two specific objectives of the thesis aimed to explore the correlation between different obesity indicators, and to develop an alternative approach to direct measurement to describe abdominal obesity using predictive models. Results suggested that Stunkard and Sorensen's silhouettes were closely related to body mass index (BMI) and reported weight, both at the time of the interview and in the past. The comparison of different predictive models showed that it was possible to estimate abdominal obesity relatively well (R2=0.64), more particularly waist circumference, from BMI, silhouette, and trousers size. Specific objectives 3 and 4 of the thesis aimed at examining the association between different indicators of individual obesity at different ages and adult obesity trajectories, and the risk of prostate cancer. Our results suggest a reduced risk of prostate cancer among overweight and obese people (odds ratio (OR) 0.71, 95% confidence interval (CI) 0.59 - 0.85). In contrast, recent abdominal obesity, estimated from various indices, was associated with an increased risk of high-grade prostate cancer (OR 1.33, 95% CI 1.03 – 1.71 for waist circumference ≥ 102 cm). Specific objectives 5 and 6 aimed to assess the familial recurrence risk of obesity according to the number of family members with prostate cancer and to assess the familial coaggregation of obesity and prostate cancer, independently of familial aggregations of obesity and prostate cancer themselves. The familial recurrence risk of obesity was higher when two or more cases of prostate cancer were observed in the family, than in the absence of cancer, regardless of the number of persons with obesity in the family. For families with 3 parents with obesity, the obesity recurrence risk was 0.35 (95% CI 0.32 – 0.37) for those with no cases of cancer in the family and 0.38 (95% CI 0.33 – 0.43) for those with at least 2 cases of cancer in the family. The difference in recurrence risks was even more marked when families whose index participant was a control were compared to those whose index case had a high-grade tumor at diagnosis. For families with 3 parents with obesity, the recurrence risk of obesity was 0.35 (95% CI 0.32 – 0.37) for those whose index participant was a control and 0.41 (95% CI 0.36 – 0.45) for those whose index case had a high-grade tumor at diagnosis. We did not observe familial co-aggregation between obesity and prostate cancer. However, this co-aggregation was present for cancers appearing before the age of 55 (OR 1.35; 95% CI 1.11 - 1.65). Results of this thesis reinforce the hypothesis of an important role of obesity in the development of prostate cancer. A particularly novel observation, obesity seems to be more frequent in families with several early-onset or aggressive tumours. Confirmation of a role of obesity, a modifiable risk factor, in the etiology of this very common cancer, would have substantial implications for public health.

This master’s thesis has aimed to contribute and fill the gap in existing studies and research where there has been a lack of knowledge about the relationship between credit ratings and stock recommendations. The purpose of this study is to analyse whether credit ratings from credit rating agencies affect financial analysts’ recommendations, as well as the opposite, if financial analysts’ recommendations affect credit rating agencies' credit ratings. The thesis has used quantitative methods with both panel data regressions where credit rating has been the dependent variable, as well as logistic regressions where recommendation has been the dependent variable. The data set has been based on firms Moody’s has issued credit ratings to between the years 1994 and 2016. The thesis’ results show that both the credit ratings from credit rating agencies and recommendations from financial analysts do indeed affect each other. This concludes that the two-way mirror between credit ratings and recommendations does exist.

A customers propensity to buy a certain product is a widely researched field and is applied in multiple industries. In this thesis it is showed that using binary classification on data from Scandinavian Airlines can predict their customers propensity to book a flight within the next coming seven days. A comparison between logistic regression and support vector machine is presented and logistic regression with reduced number of variables is chosen as the final model, due to it’s simplicity and accuracy. The explanatory variables contains exclusively booking history, whilst customer demographics and search history is showed to be insignificant. / En kunds benägenhet att göra ett visst köp är ett allmänt undersökt område som applicerats i flera olika branscher. I den här studien visas det att statistiska binära klassificeringsmodeller kan användas för att prediktera Scandinavian Airlines kunders benägenhet att köpa en resa de kommande sju dagarna. En jämförelse är presenterad mellan logistisk regression och stödvektormaskin och logistisk regression med reducerat antal parametrar väljs som den slutgiltiga modellen tack vare sin enkelhet och träffsäkerhet. De förklarande variablerna är uteslutande bokningshistorik medan kundens demografi och sökdata visas vara insignifikant.

信用評等制度在美國已有百年以上歷史，而我國自民國80幾年開始發展評等制度，截至目前，僅有中華信用評等公司與台灣經濟新報社兩家公司提供評等服務，而台灣經濟新報社更將金融保險業排除於評等對象之外。站在穩定市場競爭、保障消費者權益、配合監理需求，以及輔助專案投標等方面來看，市場上的確需要一套能反映產險業行業特性之評等模式。 本文以美國接受A.M.Best評等之產險公司為研究對象，運用三種統計方法：多元區別分析(Multiple Discriminant Analysis，MDA)、羅吉斯迴歸(Unordered Logistic Regression，ULR)、順序性羅吉斯迴歸(Ordered Logistic Regression，OLR)，來建構產險公司之信用評等模式。樣本選擇方面：估計樣本，選取美國1993年到1996年接受A.M.Best評等之產險公司327家；保留樣本，為1997年78筆資料。 而本文預定達成目標如下： 一、建立等級預測模型：參考Ederington(1985)所作債券等級預測模型，以獲利能力、槓桿、流動性、投資風險、準備金適足性五類指標共38個財務比率，透過三種統計模型，建構等級預測模型。 二、藉由等級預測之建立，尋找能有效區別產險公司評等等級之財務指標，並分析其影響程度。 三、力求模型公信力：無論變數選擇或權數決定，皆由統計軟體按照樣本特性選取產生，減少人為主觀判斷。 在決定研究對象之初，因考慮到國內產險公司接受評等之家數不多，且年數又太短，資料數量無法據以建立評等模式，因而決定以美國的產險公司為對象，再以台灣樣本作為保留樣本，預測之等級結果僅供參考之用。 / Three possible models of the P-L Insurers rating process are estimated and compared：1. Muitiple Discriminant Model, 2. Unordered Logistic Model, 3. Ordered Logistic Model. Each model is estimated for a sample of 327 American P-L insurance companies using the same 38 independent variables. The three estimated models are then employed to predict ratings for a holdout sample of 78 companies. The study analyzes 1993 through 1997 data for a sample of P-L insurers that acquired A.M.Best Financial strength ratings between December 31,1993, and December 31, 1997. Empirical evidence suggests that even when models with the same basic structure were compared, differences in estimation procedures resulted in quite different coefficient estimates and classifications. The muitiple discriminant model clearly outperformed the regression model, while the unordered logistic model was clearly superior to the ordered logistic model.