Nowadays, multi-label data are ubiquitous in real-world applications, in which each instance is associated with a set of labels. Multi-label learning has attracted significant attentions from researchers and plenty of algorithms have been proposed. Among those algorithms binary relevance (BR) is a widely used framework for multi-label classification. It constructs binary classifiers for each label by means of one-vs-rest style. BR approach is a simple and straight forward way of problem transformation for multilabel learning, but it ignores label correlations totally. Stacking based BR is a feasible way to tackle this problem. The key issue of stacking based BR is how to select label subset to extend the original features for each label. Existing methods of stacking based BR usually select identical label subset for all labels. It may be suboptimal as each label has its own most related label subset. In this paper, a novel stacking based method is introduced to utilize label correlations based on Pareto Optimum for improving the performance of BR. Our method builds a stack of two layers of BR classifiers. At the first layer, a group of binary classifiers are constructed, one for a label. At the second layer, for each label we employ Pareto Optimum to select most related label subset, then augments the original features by the selected label subset. The final binary classifiers for each label are constructed based on their corresponding reconstructed feature space. Comparing to other well-established stacking multi-label learning algorithms in terms of different multi-label classification criteria, experimental results on several multi-label benchmark datasets testify the superiority of the proposed methods.
In this paper, we present new definitions on distance and similarity measures between intuitionistic fuzzy sets (IFSs) by combining with hesitation degree. First, we discuss the limitations in traditional distance and similarity measures, which are caused by the neglect of hesitation degree's influence. Even though a vector-valued similarity measure was proposed, which has two components indicating similarity and hesitation aspects, it still cannot perform well in practical applications because hesitation works only when the values of similarity measures are equal. In order to overcome the limitations, we propose new definitions on hesitation, distance and similarity measures, and research some theorems which satisfy the requirements of the proposed definitions. Meanwhile, we investigate the relationships among hesitation, distance, similarity and entropy of IFSs to verify the consistency of our work and previous research. Finally, we analyse and discuss the advantages and disadvantages of the proposed similarity measure in detail, and then we apply the proposed measures (dH and SH) to deal with pattern recognition problems, and demonstrate that they outperform state-of-the-art distance and similarity measures.
Rough set has drawn great attention in recent decades, among which multi-granulation rough set (MGRS) is an arresting direction. It constructs a formal theoretical framework to solve complex problems under the circumstance of multiple binary relations. However, the fusion of multi-granulation rough set and grey system for acquiring knowledge is still a gap. Toward this end, we devise a grey multi-granulation rough set (GMGRS) by taking multiple grey relational relations into consideration under the framework of MGRS. In grey information system, the constructed grey relational relation that measures the relationship among objects can be used to further establish multiple binary relations. Based on two different approximate strategies (seeking common reserving difference and seeking common eliminating difference), two types of GMGRS are presented, respectively. After discussing several important properties of GMGRS, we discover that the properties of the proposed GMGRS are synchronous with the classical MGRS. Meanwhile, to obtain the attribute reduction under GMGRS, we reconstruct significance measure and termination criterion based on the θ-precision pessimistic GMGRS. Last but not least, theoretical studies and practical examples demonstrate that our proposed GMGRS largely enrich the MGRS theory and provide a new technique for knowledge discovery, which is practical in real-world scenarios.
The International Classification of Diseases, Version 9 (ICD-9) is often used to identify patients with specific diagnoses. However, certain conditions may not be accurate reflected by the ICD-9 codes, and diagnoses code assignments are complex time-consuming processes. Although there are existing methods for automotive disease diagnostic assignment techniques, they have limitations on the descriptiveness and interpretability of diseases based on features. More importantly, they ignored the importance of different features with respect to different diseases. To address the above-mentioned challenges, we propose a novel framework, namely IFFLC, which can select the most relevant features, learn disease-specific features for each disease, and perform multiple diagnosis codes' assignment. Specifically, we first develop feature selection based on disease information entropy to remove redundant and irrelevant features in both medical chart data and medical laboratory data. Then, we build a novel multiple diagnosis codes' classifier by learning the disease-specific features and exploring the intra-correlations between diseases. We employ an alternating direction method of multipliers to iteratively solve the related optimization problem. The extensive experiments on a real-world ICU database verify the superiority of the proposed method over state-of-the-art approaches.