Search

Abstract Imbalance classification is one of the most challenging research problems in machine learning. Techniques for two-class imbalance classification are relatively mature nowadays, yet multi-class imbalance learning is still an open problem. Moreover, the community lacks a suitable software tool that can integrate the major works in the field. In this paper, we present Multi-Imbalance, an open source software package for multi-class imbalanced data classification. It provides users with seven different categories of multi-class imbalance learning algorithms, including the latest advances in the field. The source codes and documentations for Multi-Imbalance are publicly available at https://github.com/chongshengzhang/Multi_Imbalance. [ABSTRACT FROM AUTHOR]

High utility itemset mining (HUIM) has emerged as an important research topic in data mining, with applications to retail-market data analysis, stock market prediction, and recommender systems, etc. However, there are very few empirical studies that systematically compare the performance of state-of-the-art HUIM algorithms. In this paper, we present an experimental evaluation on 10 major HUIM algorithms, using 9 real world and 27 synthetic datasets to evaluate their performance. Our experiments show that EFIM and d2HUP are generally the top two performers in running time, while EFIM also consumes the least memory in most cases. In order to compare these two algorithms in depth, we use another 45 synthetic datasets with varying parameters so as to study the influence of the related parameters, in particular the number of transactions, the number of distinct items and average transaction length, on the running time and memory consumption of EFIM and d2HUP. In this work, we demonstrate that, d2HUP is more efficient than EFIM under low minimum utility values and with large sparse datasets, in terms of running time; although EFIM is the fastest in dense real datasets, it is among the slowest algorithms in sparse datasets. We suggest that, when a dataset is very sparse or the average transaction length is large, and running time is favoured over memory consumption, d2HUP should be chosen. Finally, we compare d2HUP and EFIM with two newest algorithms, mHUIMiner and ULB-Miner, and find these two algorithms have moderate performance. This work has reference value for researchers and practitioners when choosing the most appropriate HUIM algorithm for their specific applications. [ABSTRACT FROM AUTHOR]

Current benchmark reports of classification algorithms generally concern common classifiers and their variants but do not include many algorithms that have been introduced in recent years. Moreover, important properties such as the dependency on number of classes and features and CPU running time are typically not examined. In this paper, we carry out a comparative empirical study on both established classifiers and more recently proposed ones on 71 data sets originating from different domains, publicly available at UCI and KEEL repositories. The list of 11 algorithms studied includes Extreme Learning Machine (ELM), Sparse Representation based Classification (SRC), and Deep Learning (DL), which have not been thoroughly investigated in existing comparative studies. It is found that Stochastic Gradient Boosting Trees (GBDT) matches or exceeds the prediction performance of Support Vector Machines (SVM) and Random Forests (RF), while being the fastest algorithm in terms of prediction efficiency. ELM also yields good accuracy results, ranking in the top-5 , alongside GBDT, RF, SVM, and C4.5 but this performance varies widely across all data sets. Unsurprisingly, top accuracy performers have average or slow training time efficiency. DL is the worst performer in terms of accuracy but second fastest in prediction efficiency. SRC shows good accuracy performance but it is the slowest classifier in both training and testing. [ABSTRACT FROM AUTHOR]

In recent years, semi-supervised learning on graphs has gained importance in many fields and applications. The goal is to use both partially labeled data (labeled examples) and a large amount of unlabeled data to build more effective predictive models. Deep Graph Neural Networks (GNNs) are very useful in both unsupervised and semi-supervised learning problems. As a special class of GNNs, Graph Convolutional Networks (GCNs) aim to obtain data representation through graph-based node smoothing and layer-wise neural network transformations. However, GCNs have some weaknesses when applied to semi-supervised graph learning: (1) it ignores the manifold structure implicitly encoded by the graph; (2) it uses a fixed neighborhood graph and focuses only on the convolution of a graph, but pays little attention to graph construction; (3) it rarely considers the problem of topological imbalance. To overcome the above shortcomings, in this paper, we propose a novel semi-supervised learning method called Re-weight Nodes and Graph Learning Convolutional Network with Manifold Regularization (ReNode-GLCNMR). Our proposed method simultaneously integrates graph learning and graph convolution into a unified network architecture, which also enforces label smoothing through an unsupervised loss term. At the same time, it addresses the problem of imbalance in graph topology by adaptively reweighting the influence of labeled nodes based on their distances to the class boundaries. Experiments on 8 benchmark datasets show that ReNode-GLCNMR significantly outperforms the state-of-the-art semi-supervised GNN methods. 1 1 The code is available at https://github.com/BiJingjun/ReNode-GLCNMR. [ABSTRACT FROM AUTHOR]

Learning from data with long-tailed and open-ended distributions is highly challenging. In this work, we propose OLPR , which is a new dual-stream O pen-set L ong-tailed recognition framework based on orthogonal P rototype learning and false R ejection correction. It consists of a Probabilistic Prediction Learning (PPL) branch and a Distance Metric Learning (DML) branch. The former is used to generate prediction probability for image classification. The latter learns orthogonal prototypes for each class by computing three distance losses, which are the orthogonal prototype loss among all the prototypes, the balanced Softmin distance based cross-entropy loss between each prototype and its corresponding input sample, and the adversarial loss for making the open-set space more compact. Furthermore, for open-set learning, instead of merely relying on binary decisions, we propose an Iterative Clustering Module (ICM) to categorize similar open-set samples and correct the false rejected closed-set samples simultaneously. If a sample is detected as a false rejection, i.e., a sample of the known classes is incorrectly identified as belonging to the unknown classes, we will re-classify the sample to the closest known/closed-set class. We conduct extensive experiments on ImageNet-LT, Places-LT, CIFAR-10/100-LT benchmark datasets, as well as a new long-tailed open-ended dataset that we build. Experimental results demonstrate that OLPR improves over the best competitors by up to 2.2% in terms of overall classification accuracy in closed-set settings, and up to 4% in terms of F-measure in open-set settings, which are very remarkable. [ABSTRACT FROM AUTHOR]