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.

nÃo hà / O mapeamento da poluiÃÃo de Ãleo no mar utilizando imagens de Radar de Abertura SintÃtica (SAR, do inglÃs Synthetic Aperture Radar) à uma importante Ãrea de interesse na Ãrea da vigilÃncia ambiental. Pode-se utilizar imagens SAR para extrair caracterÃsticas atravÃs de diferentes mÃtodos com o objetivo de predizer atravÃs de Redes Neurais Artificiais (RNAs) se uma regiÃo especÃfica contÃm ou nÃo uma mancha de Ãleo. O principal problema dessa abordagem à a ocorrÃncia de excessivos alarmes falsos decorrentes de erros de classificaÃÃo. Manchas de Ãleo sÃo eventos raros e a pequena disponibilidade de imagens contendo manchas à um fator limitante do desempenho dos classificadores. Este trabalho propÃe a utilizaÃÃo de mÃltiplos conjuntos de caracterÃsticas e mÃtodos de combinaÃÃo de classificadores para minimizar o nÃmero de alarmes falsos a fim de possibilitar a reduÃÃo de custos operacionais de sistemas automÃticos de deteÃÃo de manchas de Ãleo. As imagens SAR utilizadas neste trabalho nÃo estÃo limitadas a um Ãnico sistema de imageamento e diferentes conjuntos de caracterÃsticas baseados na geometria e textura das manchas foram testados. Os desempenhos de generalizaÃÃo de mÃtodos de combinaÃÃo de classificadores, como boosting e bagging, foram comparados com aqueles obtidos com classificadores individuais, como Perceptron Multi-Camadas (MLP, do inglÃs Multi-Layer Perceptron) e MÃquina de Vetor de Suporte (SVM, do inglÃs Support Vector Machine). Os resultados experimentais sugerem que a caracterizaÃÃo das manchas de Ãleo pode ser significativamente melhorada atravÃs do uso do boosting,mesmo quando poucas imagens amostrais estÃo disponÃveis. / Mapping ocean oil pollution by using Synthetic Aperture Radar (SAR) images is an important area of interest for environmental surveillance. One can make use of the SAR images to extract features by using different methods in order to predict if a specific region contains an oil spill or not using Artificial Neural Networks (ANN). A major problem in this approach is the number of false alarms due to misclassification. Oil spills are rare events and the number of available images containing spills is atistically small which is a limitation for the classifier performance. This work proposes the use of multiple feature sets and classifier combining methods to minimize the number of false alarms and thus, reduce the operational costs of automatic oil spill detection systems. The SAR images used in this work are not limited to a specific imaging system and different feature sets based on geometry and texture of the spills were tested. The generalization performances of classifier combination methods as boosting and bagging were compared with those resulting from single classifiers as Multilayer Perceptron (MLP) and Support Vector Machines (SVM). The experimental results suggest that oil spill characterization can be significantly improved using boosting even when few image samples are available and the feature sets have high dimensionality.