We propose a neural network-based system for automatically grading essays written in Swedish. Previous system either relies on laboriously crafted features extracted by human experts or are limited to essays written in English. By using different variations of Long Short-Term Memory (LSTM) networks, our system automatically learns the relation between Swedish high-school essays and their assigned score. Using all of the intermediate states from the LSTM network proved to be crucial in order to understand the essays. Furthermore, we evaluate different ways of representing words as dense vectors which ultimately have a substantial effect on the overall performance. We compare our results to the ones achieved by the first and previously only automatic essay scoring system designed for the Swedish language. Although no state-of-the-art performance is reached, indication of the potential from a neural based grading system is found.

Les phénotypes de tous les organismes vivants connus sont déterminés par les interactions compliquées entre les protéines produites dans ces organismes. La compréhension des réponses des organismes aux stimuli externes ou internes est basée sur la compréhension des interactions des protéines individuelles et des structures de ses complexes. La prédiction d'un complexe de deux ou plus protéines est le problème du domaine du docking protéine-protéine. Les algorithmes du docking ont habituellement deux étapes majeurs: recherche 6D exhaustive suivi par le scoring. Dans ce travail, nous avons contribués aux deus étapes sus indiquées. Nous avons développés le nouvel algorithme pour la recherche 6D exhaustive, HermiteFit. Cela est basé sur la décomposition des fonctions 3D en base Hermite. Nous avons implémenté cet algorithme dans le programme pour le fitting (l'ajustement des donnés) des cartes de densité électronique de résolution faible. Nous avons montrés qu'il surpasse les algorithmes existants en terme de temps par point tandis qu'il maintient la même précision du modèle sortant. Nous avons aussi développés la nouvelle approche de calculation de la fonction du scoring, qui est basé sur les arguments logique simples et qui évite la calculation ambiguë de l'état de référence. Nous avons comparés cela aux fonctions de scoring existantes avec l'aide du docking protéines-protéines benchmarks bien connues. Enfin, nous avons développés une approche permettant l'inclusion des interactions eau-protéine à la fonction du scoring et nous avons validés notre méthode pendant le CAPRI (Critical Assessment of Protein Interactions) tour 47. / The phenotype of every known living organism is determined mainly by the complicated interactions between the proteins produced in this organism. Understanding the orchestration of the organismal responses to the external or internal stimuli is based on the understanding of the interactions of individual proteins and their complexes structures. The prediction of a complex of two or more proteins is the problem of the protein-protein docking field. Docking algorithms usually have two major steps: exhaustive 6D rigid-body search followed by the scoring. In this work we made contribution to both of these steps. We developed a novel algorithm for 6D exhaustive search, HermiteFit. It is based on Hermite decomposition of 3D functions into the Hermite basis. We implemented this algorithm in the program for fitting low-resolution electron density maps. We showed that it outperforms existing algorithms in terms of time-per-point while maintaining the same output model accuracy. We also developed a novel approach to computation of a scoring function, which is based on simple logical arguments and avoids an ambiguous computation of the reference state. We compared it to the existing scoring functions on the widely used protein-protein docking benchmarks. Finally, we developed an approach to include water-protein interactions into the scoring functions and validated our method during the Critical Assessment of Protein Interactions round 47.

Esse trabalho consiste em uma sugestão de metodologia para análise de concessões de crédito a pessoas físicas a partir do estudo matemático e estatístico de informações sobre créditos concedidos no passado recente da carteira de crédito em questão, tais como os hábitos de pagamentos e variáveis cadastrais, financeiras, patrimoniais e de relacionamento com a instituição credora dos clientes analisados. A análise parte da definição da qualidade de crédito (bom ou ruim) dos créditos estudados, sendo seguida pelo estudo das variáveis dos clientes que influenciam na capacidade destes em honrar os compromissos do crédito obtido, a partir de técnicas estatísticas de agrupamento de indivíduos em categorias homogêneas de influência sobre a qualidade de crédito e estimação de coeficientes para as variáveis relevantes estatisticamente, através do método de regressão logística, gerando um modelo de fácil interpretação e implementação. O estudo sugere, também, ferramentas práticas para a verificação da qualidade do modelo estimado e de acompanhamento da performance da utilização do modelo durante os meses seguintes à sua implementação. Assim, o trabalho apresenta todos os aspectos vitais para análise de concessões de crédito, a partir de um enfoque bastante pragmático e tecnicamente viável.

No description available.

Title: Machine Learning for Credit Scoring Author: Elena Myazina Department / Institute: Department of Theoretical Computer Science and Mathematical Logic Supervisor of the master thesis: Mgr. Martin Pilát, Ph.D, Department of Theoretical Computer Science and Mathematical Logic Abstract: Credit scoring is a technique used by banks to evaluate their clients who ask for different types of loan. Its goal is to predict, whether a given client will pay their loan or not. Traditionally, mathematical models based on logistic regression are used for this task. In this thesis, we approach the problem of credit scoring from a machine learning point of view. We investigate several machine learning methods (including neural networks, random forests, support vector machines and other), and evaluate their performance for the credit scoring task on three publicly available datasets.. Keywords: machine learning, credit scoring,logistic regression, neural networks, random forest

In my thesis I have analyzed internal and external scoring model of financial organization. I have prepared comprehensive comparison and evaluation of both internal and external scoring systems. The aim of the thesis was creating a complete assessment of external scoring system with the simplified financial analysis and also with taking into the consideration appropriateness of this offer before approving purchase of external model.

My thesis is focused on the presentation of a scoring model for companies in chemical industry with use of cluster analysis methods. There is a description of financial risks, financial analysis indicators and models which are used to evaluate financial risks of a company. There is also a mathematical description of hierarchical cluster methods.

Ingeniero Civil Industrial / El siguiente trabajo busca establecer una metodología de seguimiento, aplicable a los modelos de riesgo crediticio de Banco Estado Microempresas (BEME), basados en regresiones logísticas, esto con el fin de levantar alertas sobre variables importantes de los modelos que están influyendo en la pérdida de poder predictivo en el tiempo. Por otro lado, se busca establecer una medida de riesgo de las pérdidas potenciales para los modelos, basadas en la conocida medida Value at Risk (VaR), con el fin de poder comparar los modelos sin recalibrar con los modelos hipotéticos de una recalibración dinámica de los mismos, capturando de forma objetiva, cambios estructurales. Para estudiar el problema de seguimiento, se busca generar una metodología que pueda ser replicable para mantener un seguimiento periódico. Para esto, se desarrolló una metodología capaz de generar de forma automática bases analíticas basada en los filtros conocidos que BEME utilizó para la creación del modelo Ambiental, el que tiene como función otorgar un puntaje a personas naturales para la pre-aprobación de un crédito. Luego, se realizó diferentes test estadísticos, en los cuales se establece un intervalo en el cual el estadístico puede oscilar, considerando que si sale de los límites establecidos, se está en presencia de cambios en las variables. Entre las pruebas utilizadas están: Beta-1, Beta-1 modificado y Fieller, los cuales mediante re-calibraciones temporales son capaces de determinar si las variables de los modelos siguen siendo de igual forma significativas. Como resultado de las pruebas, se obtuvo que para este modelo en particular la forma de calcular el criterio de bondad, que determina si se espera que será un bueno o mal cliente, representa una limitante, ya que solo es posible realizar un seguimiento a clientes con al menos un año de historial. Por otro lado para aprovechar esto se consideraron ventanas móviles de un año de la base analítica, como entrada de dato, con el fin de realizar pruebas de seguimiento más robustas y se comparó con ventanas de menor tamaño de nueve, seis y tres meses, donde se cumplió la hipótesis inicial que los test muestran mayor inestabilidad al considerar bases más pequeñas. Por último, las medidas de riesgo utilizadas mostraron resultados positivos, ya que el riesgo disminuye al re-estimar los parámetros del modelo ambiental, teniendo una incidencia de disminuir la peor perdida en un 5% del capital expuesto por el banco mensualmente en el segmento evaluado por el modelo. / 07/12/2021

This dissertation investigates the performance of two-class classi cation credit scoring data sets with low default ratios. The standard two-class parametric Gaussian and naive Bayes (NB), as well as the non-parametric Parzen classi ers are extended, using Bayes' rule, to include either a class imbalance or a Bernoulli prior. This is done with the aim of addressing the low default probability problem. Furthermore, the performance of Parzen classi cation with Silverman and Minimum Leave-one-out Entropy (MLE) Gaussian kernel bandwidth estimation is also investigated. It is shown that the non-parametric Parzen classi ers yield superior classi cation power. However, there is a longing for these non-parametric classi ers to posses a predictive power, such as exhibited by the odds ratio found in logistic regression (LR). The dissertation therefore dedicates a section to, amongst other things, study the paper entitled \Model-Free Objective Bayesian Prediction" (Bernardo 1999). Since this approach to Bayesian kernel density estimation is only developed for the univariate and the uncorrelated multivariate case, the section develops a theoretical multivariate approach to Bayesian kernel density estimation. This approach is theoretically capable of handling both correlated as well as uncorrelated features in data. This is done through the assumption of a multivariate Gaussian kernel function and the use of an inverse Wishart prior. / Dissertation (MSc)--University of Pretoria, 2017. / The financial assistance of the National Research Foundation (NRF) towards this research is hereby acknowledged. Opinions expressed and conclusions arrived at, are those of the authors and are not necessarily to be attributed to the NRF. / Statistics / MSc / Unrestricted

Determining which students are at-risk of poorer outcomes -- such as dropping out, failing classes, or decreasing standardized examination scores -- has become an important area of both research and practice in K-12 education. The models produced from this type of predictive modeling research are increasingly used by high schools in Early Warning Systems to identify which students are at risk and intervene to support better outcomes. It has become common practice to re-build and validate these detectors, district-by-district, due to different data semantics and various risk factors for students in different districts. As these detectors become more widely used, however, a new challenge emerges in applying these detectors across a broad spectrum of school districts with varying availability of past student data. Some districts have insufficient high-quality past data for building an effective detector. Novel approaches that can address the complex data challenges a new district presents are critical for advancing the field. Using an ensemble-based algorithm, I develop a modeling approach that can generate a useful model for a previously unseen district. During the ensembling process, my approach, District Similarity Ensemble Extrapolation (DSEE), weights districts that are more similar to the Target district more strongly during ensembling than less similar districts. Using this approach, I can predict student-at-risk status effectively for unseen districts, across a range of grade ranges, and achieve prediction goodness but ultimately fails to perform better than the previously published Knowles (2015) and Bowers (2012) EWS models proposed for use across districts.