Search

Graph-based semi-supervised learning has recently proved to be a powerful paradigm for processing and mining large datasets. The main advantage relies on the fact that these methods can be useful in propagating a small set of known labels to a large set of unlabeled data. However, the lack of labeled data in real applications may affect the semi-supervised learning's performance. This paper introduces a new semi-supervised framework for simultaneous linear feature extraction and label propagation. The proposed method simultaneously estimates a discriminant transformation, as well as the unknown label, by exploiting both labeled and unlabeled data. The learning model's unknowns are to be estimated by integrating two types of graph-based smoothness constraints. The resulting semi-supervised model is expected to learn more discriminative information. Experiments are conducted on nine public image datasets. These experimental results show that the proposed method's performance can be better than that of many state-of-the-art graph-based semi-supervised algorithms. [ABSTRACT FROM AUTHOR]

The advent of Big Data has ushered a new era of scientific breakthroughs. One of the common issues that affects raw data is class imbalance problem which refers to imbalanced distribution of values of the response variable. This issue is present in fraud detection, network intrusion detection, medical diagnostics, and a number of other fields where negatively labeled instances significantly outnumber positively labeled instances. Modern machine learning techniques struggle to deal with imbalanced data by focusing on minimizing the error rate for the majority class while ignoring the minority class. The goal of our paper is demonstrate the effects of class imbalance on classification models. Concretely, we study the impact of varying class imbalance ratios on classifier accuracy. By highlighting the precise nature of the relationship between the degree of class imbalance and the corresponding effects on classifier performance we hope to help researchers to better tackle the problem. To this end, we carry out extensive experiments using 10-fold cross validation on a large number of datasets. In particular, we determine that the relationship between the class imbalance ratio and the accuracy is convex. [ABSTRACT FROM AUTHOR]

Highlights • We have tackled the problem of noise in Big Data classification. • Two Big Data preprocessing approaches to remove noisy examples are proposed: HME-BD and HTE-BD. • HME-BD is a homogeneous ensemble filter, while HTE-BD is a heterogeneous ensemble filter. • The proposed algorithms constitute a first suitable noise filtering approach in Big Data domains. Abstract In any knowledge discovery process the value of extracted knowledge is directly related to the quality of the data used. Big Data problems, generated by massive growth in the scale of data observed in recent years, also follow the same dictate. A common problem affecting data quality is the presence of noise, particularly in classification problems, where label noise refers to the incorrect labeling of training instances, and is known to be a very disruptive feature of data. However, in this Big Data era, the massive growth in the scale of the data poses a challenge to traditional proposals created to tackle noise, as they have difficulties coping with such a large amount of data. New algorithms need to be proposed to treat the noise in Big Data problems, providing high quality and clean data, also known as Smart Data. In this paper, two Big Data preprocessing approaches to remove noisy examples are proposed: an homogeneous ensemble and an heterogeneous ensemble filter, with special emphasis in their scalability and performance traits. The obtained results show that these proposals enable the practitioner to efficiently obtain a Smart Dataset from any Big Data classification problem. [ABSTRACT FROM AUTHOR]

In this paper, we present a self-organising nonparametric fuzzy rule-based classifier. The proposed approach identifies prototypes from the observed data through an offline training process and uses them to build a 0-order AnYa type fuzzy rule-based system for classification. Once primed offline, it is able to continuously learn from the streaming data afterwards to follow the changing data pattern by updating the system structure and meta-parameters recursively. The meta-parameters of the proposed approach are derived from data directly. By changing the level of granularity, the proposed approach can make a trade-off between performance and computational efficiency, and, thus, the classifier is able to address a wide variety of problems with specific needs. The classifier also supports different types of distance measures. Numerical examples based on benchmark datasets demonstrate the high performance of the proposed approach and its ability of handling high-dimensional, complex, large-scale problems. [ABSTRACT FROM AUTHOR]

In real world data set the underlying data distribution may be highly skewed. Building accurate classifiers for predicting group membership is made difficult because the classifier has a tendency to be biased towards the over represented or majority group as a result. This problem is referred to as a class imbalance problem. Re-sampling techniques that produce new samples by means of over-sampling aim to combat class imbalance by increasing the number of members that belong to the minority group. This paper introduces a new over-sampling technique that focuses on noise reduction and selective sampling of the minority group which results in improvement for prediction of minority group membership. Experiments are conducted across a wide range of data sets, learners and over sampling methods. The results for this new method show improvement for Sensitivity and Gmean measures over the compared approaches. [ABSTRACT FROM AUTHOR]

The performance of a support vector machine (SVM) depends highly on the selection of the kernel function type and relevant parameters. To choose the kernel parameters properly, methods analyzing the class separability have been widely adopted for their efficiency compared with other methods, such as the popular grid search algorithm. This paper proposes a novel index called the Expected Square Distance Ratio (ESDR), which can serve as a better class separability criterion than the existing ones. Experiments on real-world datasets show that, compared with common kernel parameter selection methods that utilize the between-class separation, the variations in ESDR with respect to the kernel parameter are much more in line with those of the classification accuracy, leading to better kernel parameters. Moreover, ESDR takes the exact data distribution into account and can thus be used to study the model selection problem of an SVM for certain forms of data distribution. As an example, we employ the ESDR to analyze the selection of RBF (Radial Basis Function) kernel parameters for Gaussian data classification. [ABSTRACT FROM AUTHOR]