Class imbalance in data sets used to train classifiers can negatively affect the performance of the resulting models. A commonly used solution to address this issue is to oversample the data sets and supplement them with synthetic samples. However, oversampling can increase the level of noise in the data sets. Several sampling methods attempt to prevent this negative effect in a variety of ways, but there are still many open questions about which solutions might be appropriate in different cases. In our study, we compared the impact of different noise filtering methods on relatively small, imbalanced synthetic data sets and then examined how noise filters perform as a preprocessing step of sampling methods following different sampling strategies. The success of the noise filtering was gauged through the performance of k-Nearest Neighbours models built on the balanced data sets. Our results highlight the importance of cleaning the minority class and provide insight into which noise filtering approaches might be useful as a first step to oversampling on imbalanced noisy data sets. Among the noise filtering methods included in our study, the GMM-based ones perform well on highly noisy imbalanced data sets. It is also worth highlighting some versions of ENN, which are very effective when the noise level is moderate.

Mutual information (MI) is a widely used similarity measure in pattern recognition. MI uses entropy as a measure of uncertainty to quantify the structural similarity of two vectors. Replacing entropy with variance as a measure of uncertainty, an analogous class of similarity measures can be derived and estimated by regression techniques. Recently, the non-linear piecewise constant regression (PWCR) has been proposed to drive similarity measures of this scheme, leading to competitive alternatives of MI. Although PWCR is based on binning, the optimal binning technique for certain problems remained an open question. In this paper, we show mathematically that the optimal binning needs to be aligned with the expected relationship between the vectors being compared. In general, approximately optimal binnings can be found by combinatorial optimization, and in certain cases the optimal binning can be determined by k-means clustering. The theoretical findings are supported by numerical experiments that show a 2–5% increase in the AUC scores in simulated pattern recognition scenarios and improved feature rankings in feature selection problems. The results suggest that the proposed binning techniques could improve the performance of PWCR-driven similarity measures in real-world applications.

The openIP open source image processing library is a set of c++ libraries providing tools for education, research and industrial purposes. The aim of the development is to fill in the gap between the academic and commercial utilization of image processing. The openIP libraries are interoperable, open source and easy to install. To provide fast codes, assembler optimization, OpenMP parallelization and OpenCL based GPU utilization is integrated.