Search

It is difficult for learning models to achieve high classification performances with imbalanced data sets, because with imbalanced data sets, when one of the classes is much larger than the others, most machine learning and data mining classifiers are overly influenced by the larger classes and ignore the smaller ones. As a result, the classification algorithms often have poor learning performances due to slow convergence in the smaller classes. To balance such data sets, this paper presents a strategy that involves reducing the sizes of the majority data and generating synthetic samples for the minority data. In the reducing operation, we use the box-and-whisker plot approach to exclude outliers and the Mega-Trend-Diffusion method to find representative data from the majority data. To generate the synthetic samples, we propose a counterintuitive hypothesis to find the distributed shape of the minority data, and then produce samples according to this distribution. Four real datasets were used to examine the performance of the proposed approach. We used paired t-tests to compare the Accuracy, G-mean, and F-measure scores of the proposed data pre-processing (PPDP) method merging in the D3C method (PPDP+D3C) with those of the one-sided selection (OSS), the well-known SMOTEBoost (SB) study, and the normal distribution-based oversampling (NDO) approach, and the proposed data pre-processing (PPDP) method. The results indicate that the classification performance of the proposed approach is better than that of above-mentioned methods.

Small data-set learning problems are attracting more attention because of the short product lifecycles caused by the increasing pressure of global competition. Although statistical approaches and machine learning algorithms are widely applied to extract information from such data, these are basically developed on the assumption that training samples can represent the properties of the whole population. However, as the properties that the training samples contain are limited, the knowledge that the learning algorithms extract may also be deficient. Virtual sample generation approaches, used as a kind of data pretreatment, have proved their effectiveness when handling small data-set problems. By considering the relationships among attributes in the value generation procedure, this research proposes a non-parametric process to learn the trend similarities among attributes, and then uses these to estimate the corresponding ranges that attribute values may be located in when other attribute values are given. The ranges of the attribute values of the virtual samples are then stepwise estimated using the triangular membership functions (MFs) built to represent the attribute sample distributions. In the experiment, two real cases are examined with four modelling tools, including the M5′ model tree (M5′), multiple linear regression, support vector regression and back-propagation neural network. The results show that the forecasting accuracies of the four modelling tools are improved when training sets contain virtual samples. In addition, the outcomes of the proposed procedure show significantly smaller predictive errors than those of other approaches. [ABSTRACT FROM AUTHOR]