A hybrid method to face class overlap and class imbalance on neural networks and multi-class scenarios

Alejo, R.; Valdovinos, R. M.; García, Vicente; Pacheco-Sánchez, J. H.

doi:10.1016/j.patrec.2012.09.003

Cited by 67 publications

(31 citation statements)

References 29 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…The original GGE was proposed to improve the k-NN accuracy [15]. However, in Reference [16] the original GGE was adapted to do it effective in the backpropagation context.…”

Section: Resampling Methodsmentioning

confidence: 99%

Assessments Metrics for Multi-class Imbalance Learning: A Preliminary Study

Alejo

Antonio

Valdovinos

et al. 2013

Lecture Notes in Computer Science

Self Cite

View full text Add to dashboard Cite

Abstract. In this paper we study some of the most common global measures employed to measure the classifier performance on the multi-class imbalanced problems. The aim of this work consists of showing the relationship between global classifier performance (measure by global measures) and partial classifier performance, i.e., to determine if the results of global metrics match with the improved classifier performance over the minority classes. We have used five strategies to deal with the class imbalance problem over five real multi-class datasets on neural networks context.

show abstract

“…The original GGE was proposed to improve the k-NN accuracy [15]. However, in Reference [16] the original GGE was adapted to do it effective in the backpropagation context.…”

Section: Resampling Methodsmentioning

confidence: 99%

Assessments Metrics for Multi-class Imbalance Learning: A Preliminary Study

Alejo

Antonio

Valdovinos

et al. 2013

Lecture Notes in Computer Science

Self Cite

View full text Add to dashboard Cite

show abstract

“…However, when there is class imbalance (i.e., the non-urban samples far outnumbers the urban samples), the overall accuracy is not an appropriate evaluation criterion [74,75]. It is often biased towards the majority classes, ignoring the minority classes.…”

Section: Accessing Accuracymentioning

confidence: 99%

Mapping Regional Urban Extent Using NPP-VIIRS DNB and MODIS NDVI Data

Wang

Wan

et al. 2017

Remote Sensing

View full text Add to dashboard Cite

Abstract:The accurate and timely monitoring of regional urban extent is helpful for analyzing urban sprawl and studying environmental issues related to urbanization. This paper proposes a classification scheme for large-scale urban extent mapping by combining the Day/Night Band of the Visible Infrared Imaging Radiometer Suite on the Suomi National Polar-orbiting Partnership Satellite (NPP-VIIRS DNB) and the Normalized Difference Vegetation Index from the Moderate Resolution Imaging Spectroradiometer products (MODIS NDVI). A Back Propagation (BP) neural network based one-class classification method, the Present-Unlabeled Learning (PUL) algorithm, is employed to classify images into urban and non-urban areas. Experiments are conducted in mainland China (excluding surrounding islands) to detect urban areas in 2012. Results show that the proposed model can successfully map urban area with a kappa of 0.842 on the pixel level. Most of the urban areas are identified with a producer's accuracy of 79.63%, and only 10.42% the generated urban areas are misclassified with a user's accuracy of 89.58%. At the city level, among 647 cities, only four county-level cities are omitted. To evaluate the effectiveness of the proposed scheme, three contrastive analyses are conducted: (1) comparing the urban map obtained in this paper with that generated by the Defense Meteorological Satellite Program/Operational Linescan System Nighttime Light Data (DMSP/OLS NLD) and MODIS NDVI and with that extracted from MCD12Q1 in MODIS products; (2) comparing the performance of the integration of NPP-VIIRS DNB and MODIS NDVI with single input data; and (3) comparing the classification method used in this paper (PUL) with a linear method (Large-scale Impervious Surface Index (LISI)). According to our analyses, the proposed classification scheme shows great potential to map regional urban extents in an effective and efficient manner.

show abstract

“…Researchers in the class imbalance problem have shown their interest in finding the best samples to build the classifiers, for example eliminating those samples with a high probability to be noise or overlapped samples [18,[36][37][38][39][40], or focusing on those close to the borderline decision [11,13,41] (the latter has been less explored).…”

Section: Selective Dynamic Sampling Approachmentioning

confidence: 99%

“…• Border or overlapped samples are those samples located where the decision boundary regions intersect [18,38].…”

Section: Selective Dynamic Sampling Approachmentioning

confidence: 99%

“…In the state-of-the-art class imbalance problem, the over-and under-sampling methods are very popular and successful approaches to deal with this problem (see [7,8,10,[16][17][18][19]). Over-sampling replicates samples in the minority-class, and under-sampling eliminates samples from the majority-class, biasing the discrimination process to compensate for the class imbalance.…”

Section: State-of-the-art Of the Class Imbalance Approachesmentioning

confidence: 99%

See 1 more Smart Citation

A Selective Dynamic Sampling Back-Propagation Approach for Handling the Two-Class Imbalance Problem

Alejo¹,

Monroy-de-Jesús

Pacheco-Sánchez

et al. 2016

Applied Sciences

View full text Add to dashboard Cite

Abstract:In this work, we developed a Selective Dynamic Sampling Approach (SDSA) to deal with the class imbalance problem. It is based on the idea of using only the most appropriate samples during the neural network training stage. The "average samples"are the best to train the neural network, they are neither hard, nor easy to learn, and they could improve the classifier performance. The experimental results show that the proposed method is a successful method to deal with the two-class imbalance problem. It is very competitive with respect to well-known over-sampling approaches and dynamic sampling approaches, even often outperforming the under-sampling and standard back-propagation methods. SDSA is a very simple method for automatically selecting the most appropriate samples (average samples) during the training of the back-propagation, and it is very efficient. In the training stage, SDSA uses significantly fewer samples than the popular over-sampling approaches and even than the standard back-propagation trained with the original dataset.

show abstract

A hybrid method to face class overlap and class imbalance on neural networks and multi-class scenarios

Cited by 67 publications

References 29 publications

Assessments Metrics for Multi-class Imbalance Learning: A Preliminary Study

Assessments Metrics for Multi-class Imbalance Learning: A Preliminary Study

Mapping Regional Urban Extent Using NPP-VIIRS DNB and MODIS NDVI Data

A Selective Dynamic Sampling Back-Propagation Approach for Handling the Two-Class Imbalance Problem

Contact Info

Product

Resources

About