PRISM ? A novel framework for pattern recognition

Effect of label noise in the complexity of classification problems

“…Singh [22] recommends a technique that estimates the complexity of the classification problem using neighborhood information for the identification of outliers. Sáez et al [23] use measures able to characterize the complexity of the classification problem to predict when a noise filter can be effectively applied to a dataset.…”

A meta-learning framework for pattern classification by means of data complexity measures

“…In these situations, a number of statistical probability distances such as Bhattacharya, Chernoff, Mahalanobis, Matusita, etc. provide upper and lower bounds for the error as a special case for a two-class problem [11].…”

“…Singh [11] employs several data complexity measures to remove outliers from a training set and also points out that another utility would be to help reduce the data set size without compromising on the test performance of classifiers. More specifically, those patterns that are found deep inside class boundaries could be removed from the training set since they are least likely to help in classification of test samples.…”

“…Various recent papers [1,6,11,13] have introduced the use of measures to characterize the data complexity and to relate such descriptions to classifier performance.…”

“…For example, Singh [11] conducts an empirical study to demonstrate that the employment of a neighborhood separability measure constitutes a suitable criterion for optimization in feature selection for a face recognition problem.…”