Flexible multi-class cost-sensitive thresholding

There is growing interest in multilabel image classification due to its critical role in web-based image analytics-based applications, such as large-scale image retrieval and browsing. Matrix completion (MC) has recently been introduced as a method for transductive (semisupervised) multilabel classification, and has several distinct advantages, including robustness to missing data and background noise in both feature and label space. However, it is limited by only considering data represented by a single-view feature, which cannot precisely characterize images containing several semantic concepts. To utilize multiple features taken from different views, we have to concatenate the different features as a long vector. However, this concatenation is prone to over-fitting and often leads to very high time complexity in MC-based image classification. Therefore, we propose to weightedly combine the MC outputs of different views, and present the multiview MC (MVMC) framework for transductive multilabel image classification. To learn the view combination weights effectively, we apply a cross-validation strategy on the labeled set. In particular, MVMC splits the labeled set into two parts, and predicts the labels of one part using the known labels of the other part. The predicted labels are then used to learn the view combination coefficients. In the learning process, we adopt the average precision (AP) loss, which is particular suitable for multilabel image classification, since the ranking-based criteria are critical for evaluating a multilabel classification system. A least squares loss formulation is also presented for the sake of efficiency, and the robustness of the algorithm based on the AP loss compared with the other losses is investigated. Experimental evaluation on two real-world data sets (PASCAL VOC' 07 and MIR Flickr) demonstrate the effectiveness of MVMC for transductive (semisupervised) multilabel image classification, and show that MVMC can exploit complementary properties of different features and output-consistent labels for improved multilabel image classification.

New variants of pairwise classification

Automated classification of prostate histopathology images includes the identification of multiple classes, such as benign and cancerous (grades 3 & 4). To address the multiclass classification problem in prostate histopathology images, breakdown approaches are utilized, such as one-versus-one (OVO) and one- versus-all (Ovall). In these approaches, the multiclass problem is decomposed into numerous binary subtasks, which are separately addressed. However, OVALL introduces an artificial class imbalance, which degrades the classification performance, while in the case of OVO, the correlation between different classes not regarded as a multiclass problem is broken into multiple independent binary problems. This paper proposes a new multiclass approach called multi-level (hierarchical) learning architecture (MLA). It addresses the binary classification tasks within the framework of a hierarchical strategy. It does so by accounting for the interaction between several classes and the domain knowledge. The proposed approach relies on the ‘divide-and-conquer’ principle by allocating each binary task into two separate subtasks; strong and weak, based on the power of the samples in each binary task. Conversely, the strong samples include more information about the considered task, which motivates the production of the final prediction. Experimental results on prostate histopathological images illustrated that the MLA significantly outperforms the Ovall and OVO approaches when applied to the ensemble framework. The results also confirmed the high efficiency of the ensemble framework with the MLA scheme in dealing with the multiclass classification problem.

In this paper, we propose a novel approach for obtaining predictions from per-class scores to improve the accuracy of multi-label classification systems. In a multi-label classification task, the expected output is a set of predicted labels per each testing sample. Typically, these predictions are calculated by implicit or explicit thresholding of per-class real-valued scores: classes with scores exceeding a given threshold value are added to a prediction set. In our work, we propose a neural network-based thresholding phase for multi-label classification systems and examine its influence on the overall classification performance measured by micro- and macro-averaged F1 scores on synthetic and real datasets. In contrast to classic thresholding methods, our approach has the unique property of being able to recover from scoring errors, because each decision about a given label prediction depends on the corresponding class score, as well as on all the other class scores for a given sample at once. The method can be used in combination with any classification system that outputs real-valued class scores. The proposed thresholding methods are trained offline, after the completion of the scoring phase. As such, it can be considered a universal fine-tuning step that can be employed in any multi-label classification system that seeks to find the best multi-label predictions based on class scores. In our experiments on real datasets, the input class scores were obtained from two third-party baseline classification systems. We show that our approach outperforms the traditional thresholding methods, which results in the improved performance of all tested multi-label classification tasks. In terms of relative improvement, on real datasets, the micro-F1 score is higher by up to 40.6%, the macro-F1 score is higher by up to 3.6%, and the averaged micro–macro-F1 score is higher by up to 30.1%, considering single models only. We show that ensembles and hybrid models give even better results. We show examples of successful extreme recoveries, where the system, equipped with our method, was able to correctly predict labels, which were highly underscored after the scoring phase.

A new discriminant subspace analysis approach for multi-class problems

Text categorization via generalized discriminant analysis

Multi-class imbalanced data classification problem is common in the real world, but traditional binary classification methods cannot be directly applied. Existing solutions include designing new multi-class classification algorithm and dividing multi-class classification problem into binary classification problem. The latter includes two widely used strategies, namely one versus all (OVA) and one versus one (OVO). In this paper, we propose a combination method based on all and one (A&O), which is a combination of OVA and OVO, for multi-class imbalanced data classification problem. The method is developed by combining A&O and data balancing technique named SMOTE. Comparative experiments on 13 UCI datasets show that the proposed method performs well.

Text categorization is an important research area and has been receiving much attention due to the growth of the on-line information and of Internet. Automated text categorization is generally cast as a multi-class classification problem. Much of previous work focused on binary document classification problems. Support vector machines (SVMs) excel in binary classification, but the elegant theory behind large-margin hyperplane cannot be easily extended to multi-class text classification. In addition, the training time and scaling are also important concerns. On the other hand, other techniques naturally extensible to handle multi-class classification are generally not as accurate as SVM. This paper presents a simple and efficient solution to multi-class text categorization. Classification problems are first formulated as optimization via discriminant analysis. Text categorization is then cast as the problem of finding coordinate transformations that reflects the inherent similarity from the data. While most of the previous approaches decompose a multi-class classification problem into multiple independent binary classification tasks, the proposed approach enables direct multi-class classification. By using Generalized Singular Value Decomposition (GSVD), a coordinate transformation that reflects the inherent class structure indicated by the generalized singular values is identified. Extensive experiments demonstrate the efficiency and effectiveness of the proposed approach.

Anemia is one of the most pressing public health issues in the world with iron deficiency a major public health issue worldwide. The highest prevalence of anemia is in developing countries. The complete blood count is a blood test used to diagnose the prevalence of anemia. While earlier studies have framed the problem of diagnosis as a binary classification problem, this paper frames it as a multi class (three classes) classification problem with mild, moderate and severe classes. The three classes for the anemia classification (mild, moderate, severe) are so chosen as the world health organization (WHO) guidelines formalize this categorization based on the Haemoglobin (HGB) values of the chosen sample of patients in the Complete Blood Count (CBC) patient data set. Complete blood count test data was collected in an outpatient clinical setting in India. We used Feature selection with Majority voting to identify the key attributes in the input patient data set. In addition, since the original data set was imbalanced we used Synthetic Minority Oversampling Technique (SMOTE) to balance the data set. Four data sets including the original data set were used to perform the data experiments. Six standard machine learning algorithms were utilised to test our four data sets, performing multi class classification. Benchmarking these algorithms was performed and tabulated using both10 fold cross validation and hold out methods. The experimental results indicated that multilayer perceptron network was predominantly giving good recall values across mild and moderate class which are early and middle stages of the disease. With a good prediction model at early stages, medical intervention can provide preventive measure from further deterioration into severe stage or recommend the use of supplements to overcome this problem.

Anemia is one of the most pressing public health issues in the world with iron deficiency a major public health issue worldwide. The highest prevalence of anemia is in developing countries. The complete blood count is a blood test used to diagnose the prevalence of anemia. While earlier studies have framed the problem of diagnosis as a binary classification problem, this paper frames it as a multi class (three classes) classification problem with mild, moderate and severe classes. The three classes for the anemia classification (mild, moderate, severe) are so chosen as the world health organization (WHO) guidelines formalize this categorization based on the Haemoglobin (HGB) values of the chosen sample of patients in the Complete Blood Count (CBC) patient data set. Complete blood count test data was collected in an outpatient clinical setting in India. We used Feature selection with Majority voting to identify the key attributes in the input patient data set. In addition, since the original data set was imbalanced we used Synthetic Minority Oversampling Technique (SMOTE) to balance the data set. Four data sets including the original data set were used to perform the data experiments. Six standard machine learning algorithms were utilised to test our four data sets, performing multi class classification. Benchmarking these algorithms was performed and tabulated using both10 fold cross validation and hold out methods. The experimental results indicated that multilayer perceptron network was predominantly giving good recall values across mild and moderate class which are early and middle stages of the disease. With a good prediction model at early stages, medical intervention can provide preventive measure from further deterioration into severe stage or recommend the use of supplements to overcome this problem.

Several classification problems involve more than two classes. These problems are known as multiclass classification problems. One of the approaches to deal with multiclass problems is their decomposition into a set of binary problems. Recent work shows important advantages related with this approach. Several strategies have been proposed for this decomposition. The strategies most frequently used are All-vs-All, One-vs-All and Error Correction Output Codes (ECOC). ECOCs are based on binary words (codewords) and have been adapted to deal with multiclass problems. For such, they must comply with a number of specific constraints. Different dimensions may be adopted for the codewords for each number of classes in the problem. These dimensions grow exponentially with the number of classes present in a dataset. Two methods to choose the dimension of a ECOC, which assure a good trade-off between redundancy and error correction capacity, are proposed in this paper. The proposed methods are evaluated in a set of benchmark classification problems. Experimental results show that they are competitive with other multiclass decomposition methods.

Exploiting local label correlation from sample perspective for multi-label classification via three-way decision theory

This paper presents a novel architecture based on a constructive algorithm that allows the network to grow attending to both supervised and unsupervised criteria. The main goal is to end up with a set of discriminant functions able to solve a multi-class classification problem. The main difference with well-known NN-classificators lean on the fact that training is performed over labeled sets of patterns that we call high-level-structures (HLS). Every set contain patterns linked each other by some physical evidence, like neighbor pixels in a subimage or a time-sequence of frequency vectors in a speech utterance, but the membership of every individual pattern in the high-level-structure can not be so clear. This architecture has been tested on a number of artificial data sets and real data sets with very good results. We are now applying the algorithm to classification of real images drawn from the DataBase created for the ALINSPEC project.

Cost-Sensitive back-propagation neural networks with binarization techniques in addressing multi-class problems and non-competent classifiers

In this article, we present real and synthetic data sets for benchmarking key-values stores. Here, we focus on various data types and sizes. Key-value pairs in key-value data sets consist of the key and the value. We can construct any kinds of data as key-value data sets by assigning an arbitrary type of data as the value and a unique ID as the key. Therefore, key-value pairs are quite worthy when we deal with big data because the data types in the big data application become more various and, even sometimes, they are not known or determined. In this article, we crawl four kinds of real data sets by varying the type of data sets (i.e., variety) and generate four kinds of synthetic data sets by varying the size of data sets (i.e., volume). For real data sets, we crawl data sets with various data types from Twitter, i.e., Tweets in text, a list of hashtags, geo-location of the tweet, and the number of followers. We also present algorithms for crawling real data sets based on REST APIs and streaming APIs and for generating synthetic data sets. Using those algorithms, we can crawl any key-value pairs of data types supported by Twitter and can generate any size of synthetic data sets by extending them simply. Last, we show that the crawled and generated data sets are actually utilized for the well-known key-value stores such as Level DB of Google, RocksDB of Facebook, and Berkeley DB of Oracle. Actually, the presented real and synthetic data sets have been used for comparing the performance of them. As an example, we present an algorithm of the basic operations for the key-value stores of LevelDB.