This paper presents a robust classification method, the ROC Convex Hull (ROCCH), for handling imprecise environments where target operating conditions are not precisely known. Traditional classification systems often rely on a single best classifier, but this approach is brittle when target conditions change. The ROCCH method combines techniques from ROC analysis, decision analysis, and computational geometry to build a hybrid classifier that performs at least as well as the best available classifier for any target conditions. It is efficient, robust to imprecise class distributions and misclassification costs, and allows for clear visual comparisons and sensitivity analyses. The ROCCH method is based on the idea of comparing classifier performance in ROC space, where the true positive rate (TP) is on the Y-axis and the false positive rate (FP) is on the X-axis. Each classifier is represented by a point in ROC space, and the ROC convex hull (ROCCH) identifies the subset of classifiers that are potentially optimal under any combination of cost assumptions and class distribution assumptions. The ROCCH method is efficient and incremental, minimizing the management of classifier performance data and allowing for clear visual comparisons. The paper demonstrates that the ROCCH method can be used to build a hybrid classification system that performs best under any target cost/benefit and class distributions. Target conditions can be specified at run time, and the system can be tuned easily based on feedback from its actual performance. The ROCCH method is robust across a wide variety of comparison frameworks, including the optimization of metrics such as accuracy, expected cost, lift, precision, recall, and workforce utilization. It is also efficient to build, store, and update. The paper also shows that the ROCCH method can outperform the best known classifier in certain situations. Empirical evidence supports the need for a robust hybrid classifier in many real-world problems. The ROCCH method is particularly useful in situations where the target cost distribution or class distribution is completely unknown, as it allows for the selection of a classifier that maximizes a single-number metric representing the average performance over the entire curve. The ROCCH method is also useful in ranking cases and optimizing workforce utilization. The paper concludes that the ROCCH method provides a robust solution for classification in imprecise environments.This paper presents a robust classification method, the ROC Convex Hull (ROCCH), for handling imprecise environments where target operating conditions are not precisely known. Traditional classification systems often rely on a single best classifier, but this approach is brittle when target conditions change. The ROCCH method combines techniques from ROC analysis, decision analysis, and computational geometry to build a hybrid classifier that performs at least as well as the best available classifier for any target conditions. It is efficient, robust to imprecise class distributions and misclassification costs, and allows for clear visual comparisons and sensitivity analyses. The ROCCH method is based on the idea of comparing classifier performance in ROC space, where the true positive rate (TP) is on the Y-axis and the false positive rate (FP) is on the X-axis. Each classifier is represented by a point in ROC space, and the ROC convex hull (ROCCH) identifies the subset of classifiers that are potentially optimal under any combination of cost assumptions and class distribution assumptions. The ROCCH method is efficient and incremental, minimizing the management of classifier performance data and allowing for clear visual comparisons. The paper demonstrates that the ROCCH method can be used to build a hybrid classification system that performs best under any target cost/benefit and class distributions. Target conditions can be specified at run time, and the system can be tuned easily based on feedback from its actual performance. The ROCCH method is robust across a wide variety of comparison frameworks, including the optimization of metrics such as accuracy, expected cost, lift, precision, recall, and workforce utilization. It is also efficient to build, store, and update. The paper also shows that the ROCCH method can outperform the best known classifier in certain situations. Empirical evidence supports the need for a robust hybrid classifier in many real-world problems. The ROCCH method is particularly useful in situations where the target cost distribution or class distribution is completely unknown, as it allows for the selection of a classifier that maximizes a single-number metric representing the average performance over the entire curve. The ROCCH method is also useful in ranking cases and optimizing workforce utilization. The paper concludes that the ROCCH method provides a robust solution for classification in imprecise environments.