One of the most mentioned credit risk parameters in banking sector is loss given default (LGD). The regulatory framework allows to use own LGD estimation procedures after approval. The classification and regression trees are appropriate and flexible in this context and they offer some advantages comparing to the traditional approaches such as linear regression model. This work includes a theoretical background on tree based methods. In the last section, loss given default from debit accounts is estimated using the random forests which show the best performance in this case.

No description available.

abstract: Longitudinal recursive partitioning (LRP) is a tree-based method for longitudinal data. It takes a sample of individuals that were each measured repeatedly across time, and it splits them based on a set of covariates such that individuals with similar trajectories become grouped together into nodes. LRP does this by fitting a mixed-effects model to each node every time that it becomes partitioned and extracting the deviance, which is the measure of node purity. LRP is implemented using the classification and regression tree algorithm, which suffers from a variable selection bias and does not guarantee reaching a global optimum. Additionally, fitting mixed-effects models to each potential split only to extract the deviance and discard the rest of the information is a computationally intensive procedure. Therefore, in this dissertation, I address the high computational demand, variable selection bias, and local optimum solution. I propose three approximation methods that reduce the computational demand of LRP, and at the same time, allow for a straightforward extension to recursive partitioning algorithms that do not have a variable selection bias and can reach the global optimum solution. In the three proposed approximations, a mixed-effects model is fit to the full data, and the growth curve coefficients for each individual are extracted. Then, (1) a principal component analysis is fit to the set of coefficients and the principal component score is extracted for each individual, (2) a one-factor model is fit to the coefficients and the factor score is extracted, or (3) the coefficients are summed. The three methods result in each individual having a single score that represents the growth curve trajectory. Therefore, now that the outcome is a single score for each individual, any tree-based method may be used for partitioning the data and group the individuals together. Once the individuals are assigned to their final nodes, a mixed-effects model is fit to each terminal node with the individuals belonging to it. I conduct a simulation study, where I show that the approximation methods achieve the goals proposed while maintaining a similar level of out-of-sample prediction accuracy as LRP. I then illustrate and compare the methods using an applied data. / Dissertation/Thesis / Doctoral Dissertation Psychology 2019

No description available.

An elasmobranch survey conducted from 2013-2018 in the waters adjacent to Pinellas County, Florida, was used for a baseline assessment of the local shark population. ArcGIS and Boosted Regression Trees were used to identify hot spots of abundance and links between environmental predictors and distribution, as well as create species distribution models. A diverse assemblage of sharks, dominated by five species: nurse shark, bonnethead, Atlantic sharpnose shark, blacktip shark, and blacknose shark, was identified. A large proportion of captures (~42%) were immature sharks. Results indicate areas characterized by seagrass and “No Internal Combustion Engine” zones correlate with greater diversity and abundance, particularly for immature sharks. BRT results underscored the importance of seagrass bottoms, as well as warm (>31℃) and shallow (< 6m) waters as essential habitat. By identifying spatially explicit areas and environmental conditions suited for shark abundance, this study provides practical resources for managing and protecting Florida’s sharks.

No description available.

No description available.

<p>In recent years, measuring customer satisfaction has become one of the key concerns of market research studies. One of the basic features of leading companies is their success in fulfilling their customers’ demands. For that reason, companies attempt to find out what essential factors dominate their customers’ purchasing habits.</p> <p>Millennium Research Group (MRG) - a global authority on medical tech- nology market intelligence - uses a web-based survey tool to collect informa- tion about customers’ level of satisfaction. One of their surveys is designed to gather information about the practitioner’s level of satisfaction on different brands of dental implants. The Dental Implants dataset obtained from the survey tool has thirty-four attributes, and practitioners were asked to rank or specify their level of satisfaction by assigning a score to each attribute.</p> <p>The basic question asked by the company was whether the attributes were useful to make customer behavior predictions. The aim of this study is to assess the reliability and accuracy of these measures and to build a model for future predictions, then, determine the attributes that are most influential</p> <p>in the practitioners’ purchasing decisions. Classification and regression trees (CART) and Partial least squares regression (PLSR) are the two statistical approaches used in this study to build a prediction model for the Dental Implants dataset.</p> <p>The prediction models generated, using both of the techniques, have rel- atively small prediction powers; which may be perceived as an indication of deficiency in the dataset. However, getting a small prediction power is gener- ally expected in market research studies. The research then attempts to find ways to improve the power of these models to get more accurate results. The model generated by CART analysis tends to have better prediction power and is more suitable for future predictions. Although PLSR provides extremely small prediction power, it helps finding out the most important attributes that influence the practitioners’ purchasing decisions. Improvements in pre- diction are sought by restricting the cases in the data to subsets that show better alignment between predictors and customer purchasing behaviour.</p> / Master of Science (MSc)

Most estimates of wood supply do not reflect the true availability of wood resources. The availability of wood resources ultimately depends on collective wood harvesting decisions across the landscape. Both social and biophysical constraints impact harvesting decisions and thus the availability of wood resources. While most constraints do not completely inhibit harvesting, they may significantly reduce the probability of harvest. Realistic assessments of woody availability and distribution are needed for effective forest management and planning. This study focuses on predicting the probability of harvest at forested FIA plot locations in Virginia. Classification and regression trees, conditional inferences trees, random forest, balanced random forest, conditional random forest, and logistic regression models were built to predict harvest as a function of social and biophysical availability constraints. All of the models were evaluated and compared to identify important variables constraining harvest, predict future harvests, and estimate the available wood supply. Variables related to population and resource quality seem to be the best predictors of future harvest. The balanced random forest and logistic regressions models are recommended for predicting future harvests. The balanced random forest model is the best predictor, while the logistic regression model can be most easily shared and replicated. Both models were applied to predict harvest at recently measured FIA plots. Based on the probability of harvest, we estimate that between 2012 and 2017, 10 – 21 percent of total wood volume on timberland will be available for harvesting. / Master of Science

With exploding human populations and landscapes that are changing, an increasing number of wildlife species are brought to the brink of extinction. In Canada, the eastern hog-nosed snake, 'Heterodon platirhinos', is found in a limited portion of southern Ontario. Designated as threatened by the Committee on the Status of Endangered Wildlife in Canada (COSEWIC), this reptile has been losing its habitat at an alarming rate. Due to the increase in development of southern Ontario, it is crucial to document what limits the snake’s habitat to direct conservation efforts better, for the long-term survival of this species. The goals of this study are: 1) to examine what environmental parameters are linked to the presence of the species at a landscape scale; 2) to predict where the snakes can be found in Ontario through GIS-based habitat suitability models (HSMs); and 3) to assess the role of biotic interactions in HSMs. Three models with high predictive power were employed: Maxent, Boosted Regression Trees (BRTs), and the Genetic Algorithm for Rule-set Production (GARP). Habitat suitability maps were constructed for the eastern hog-nosed snake for its entire Canadian distribution and models were validated with both threshold dependent and independent metrics. Maxent and BRT performed better than GARP and all models predict fewer areas of high suitability when landscape variables are used with current occurrences. Forest density and maximum temperature during the active season were the two variables that contributed the most to models predicting the current distribution of the species. Biotic variables increased the performance of models not by representing a limiting resource, but by representing the inequality of sampling and areas where forest remains. Although habitat suitability models rely on many assumptions, they remain useful in the fields of conservation and landscape management. In addition to help identify critical habitat, HSMs may be used as a tool to better manage land to allow for the survival of species at risk.