Currently, a lack of accurate emission data exits for particulate matter (PM) in agricultural air quality studies (USDA-AAQTF, 2000). PM samplers, however, tend to over estimate the concentration of most agricultural dusts because of the interaction of the particle size distribution (PSD) and performance characteristics of the sampler (Buser, 2004). This research attempts to find a practical method to characterize and correct this error for the Federal Reference Method (FRM) PM10 sampler. First, a new dust wind tunnel testing facility that satisfies the USEPA’s requirement of testing PM10 samplers was designed, built, and evaluated. Second, the wind tunnel testing protocol using poly-dispersed aerosol as the test dust was proved to be able to provide results consistent with mono-dispersed dusts. Third, this study quantified the variation of over sampling ratios for the various cut point and slopes of FRM PM10 samplers and proposed an averaged over sampling ratio as a correction factor for various ranges of PSD. Finally, a method of using total suspended particle (TSP) samplers as a field reference for determining PM10 concentrations and aerosol PSD was explored computationally. Overall, this dissertation developed successfully the methodology to correct the sampling error associated with the FRM PM10 sampler: (1) wind tunnel testing facilities and protocol for experimental evaluation of samplers; (2) the variation of the oversampling ratios of FRM PM10 samplers for computational evaluation of samplers; (3) the evaluation of TSP sampler effectiveness as a potential field reference for field evaluation of samplers.

Existing classification analysis techniques (e.g., decision tree induction,) generally exhibit satisfactory classification effectiveness when dealing with data with non-skewed class distribution. However, real-world applications (e.g., churn prediction and fraud detection) often involve highly skewed data in decision outcomes. Such a highly skewed class distribution problem, if not properly addressed, would imperil the resulting learning effectiveness. In this study, we empirically evaluate three different approaches, namely the under-sampling, the over-sampling and the multi-classifier committee approaches, for addressing classification with highly skewed class distribution. Due to its popularity, C4.5 is selected as the underlying classification analysis technique. Based on 10 highly skewed class distribution datasets, our empirical evaluations suggest that the multi-classifier committee generally outperformed the under-sampling and the over-sampling approaches, using the recall rate, precision rate and F1-measure as the evaluation criteria. Furthermore, for applications aiming at a high recall rate, use of the over-sampling approach will be suggested. On the other hand, if the precision rate is the primary concern, adoption of the classification model induced directly from original datasets would be recommended.

Two approaches to building models for prediction of the onset of Type 1 diabetes mellitus in juvenile subjects were examined. A set of tests performed immediately before diagnosis was used to build classifiers to predict whether the subject would be diagnosed with juvenile diabetes. A modified training set consisting of differences between test results taken at different times was also used to build classifiers to predict whether a subject would be diagnosed with juvenile diabetes. Neural networks were compared with decision trees and ensembles of both types of classifiers. Support Vector Machines were also tested on this dataset. The highest known predictive accuracy was obtained when the data was encoded to explicitly indicate missing attributes in both cases. In the latter case, high accuracy was achieved without test results which, by themselves, could indicate diabetes. The effects of oversampling of minority class samples in the training set by generating synthetic examples were tested with ensemble techniques like bagging and random forests. It was observed, that oversampling of diabetic examples, lead to an increased accuracy in diabetic prediction demonstrated by a significantly better F-measure value. ROC curves and the statistical F-measure were used to compare the performance of the different machine learning algorithms.

Submitted by Gabriel Lento (gabriel.carneiro.lento@gmail.com) on 2017-05-01T23:16:04Z No. of bitstreams: 1 Dissertação Gabriel Carneiro Lento.pdf: 832965 bytes, checksum: f79e7cb4e5933fd8c3a7c67ed781ddb5 (MD5) / Approved for entry into archive by Leiliane Silva (leiliane.silva@fgv.br) on 2017-05-04T18:39:57Z (GMT) No. of bitstreams: 1 Dissertação Gabriel Carneiro Lento.pdf: 832965 bytes, checksum: f79e7cb4e5933fd8c3a7c67ed781ddb5 (MD5) / Made available in DSpace on 2017-05-17T12:43:35Z (GMT). No. of bitstreams: 1 Dissertação Gabriel Carneiro Lento.pdf: 832965 bytes, checksum: f79e7cb4e5933fd8c3a7c67ed781ddb5 (MD5) Previous issue date: 2017-03-27 / In this work we study churn in health insurance, that is predicting which clients will cancel the product or service within a preset time-frame. Traditionally, the probability whether a client will cancel the service is modeled using logistic regression. Recently, modern machine learning techniques are becoming popular in churn modeling, having been applied in the areas of telecommunications, banking, and car insurance, among others. One of the big challenges in this problem is that only a fraction of all customers cancel the service, meaning that we have to deal with highly imbalanced class probabilities. Under-sampling and over-sampling techniques have been used to overcome this issue. We use random forests, that are ensembles of decision trees, where each of the trees fits a subsample of the data constructed using either under-sampling or over-sampling. We compare the distinct specifications of random forests using various metrics that are robust to imbalanced classes, both in-sample and out-of-sample. We observe that random forests using imbalanced random samples with fewer observations than the original series present a better overall performance. Random forests also present a better performance than the classical logistic regression, often used in health insurance companies to model churn. / Neste trabalho estudamos o problema de churn em seguro saúde, isto é, a previsão se o cliente irá cancelar o produto ou serviço em até um período de tempo pré-estipulado. Tradicionalmente, regressão logística é utilizada para modelar a probabilidade de cancelamento do serviço. Atualmente, técnicas modernas de machine learning vêm se tornando cada vez mais populares para esse tipo de problema, com exemplos nas áreas de telecomunicação, bancos, e seguros de carro, dentre outras. Uma das grandes dificuldades nesta modelagem é que apenas uma pequena fração dos clientes de fato cancela o serviço, o que significa que a base de dados tratada é altamente desbalanceada. Técnicas de under-sampling e over-sampling são utilizadas para contornar esse problema. Neste trabalho, aplicamos random forests, que são combinações de árvores de decisão ajustadas em subamostras dos dados, construídas utilizando under-sampling e over-sampling. Ao fim do trabalho comparamos métricas de ajustes obtidas nas diversas especificações dos modelos testados e avaliamos seus resultados dentro e fora da amostra. Observamos que técnicas de random forest utilizando sub-amostras não balanceadas com o tamanho menor do que a amostra original apresenta a melhor performance dentre as random forests utilizadas e uma melhora com relação ao praticado no mercado de seguro saúde.

In this thesis, the performance of two over-sampling techniques, SMOTE and ADASYN, is compared. The comparison is done on three imbalanced data sets using three different classification models and evaluation metrics, while varying the way the data is pre-processed. The results show that both SMOTE and ADASYN improve the performance of the classifiers in most cases. It is also found that SVM in conjunction with SMOTE performs better than with ADASYN as the degree of class imbalance increases. Furthermore, both SMOTE and ADASYN increase the relative performance of the Random forest as the degree of class imbalance grows. However, no pre-processing method consistently outperforms the other in its contribution to better performance as the degree of class imbalance varies.

[pt] O objetivo deste estudo foi explorar o problema de churn em seguros de vida, no sentido de prever se o cliente irá cancelar o produto nos próximos 6 meses. Atualmente, métodos de machine learning vêm se popularizando para este tipo de análise, tornando-se uma alternativa ao tradicional método de modelagem da probabilidade de cancelamento através da regressão logística. Em geral, um dos desafios encontrados neste tipo de modelagem é que a proporção de clientes que cancelam o serviço é relativamente pequena. Para isso, este estudo recorreu a técnicas de balanceamento para tratar a base naturalmente desbalanceada – técnicas de undersampling, oversampling e diferentes combinações destas duas foram utilizadas e comparadas entre si. As bases foram utilizadas para treinar modelos de Bagging, Random Forest e Boosting, e seus resultados foram comparados entre si e também aos resultados obtidos através do modelo de Regressão Logística. Observamos que a técnica SMOTE-modificado para balanceamento da base, aplicada ao modelo de Bagging, foi a combinação que apresentou melhores resultados dentre as combinações exploradas. / [en] The purpose of this study is to explore the churn problem in life insurance, in the sense of predicting if the client will cancel the product in the next 6 months. Currently, machine learning methods are becoming popular in this type of analysis, turning it into an alternative to the traditional method of modeling the probability of cancellation through logistics regression. In general, one of the challenges found in this type of modelling is that the proportion of clients who cancelled the service is relatively small. For this, the study resorted to balancing techniques to treat the naturally unbalanced base – under-sampling and over-sampling techniques and different combinations of these two were used and compared among each other. The bases were used to train models of Bagging, Random Forest and Boosting, and its results were compared among each other and to the results obtained through the Logistics Regression model. We observed that the modified SMOTE technique to balance the base, applied to the Bagging model, was the combination that presented the best results among the explored combinations.