Effect of label noise in the complexity of classification problems

García, L.; Carvalho, André C. P. L. F. de; Lorena, Ana Carolina

doi:10.1016/j.neucom.2014.10.085

Cited by 120 publications

(52 citation statements)

References 28 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…For example, the average path length of training instances may increase in the presence of class label noise, leading to increasing computational training time. In order to evaluate the impact of class label noise on the classifier complexity, several measures, such as the class separability, have been studied in [38]. The effect of noise presence has also been evaluated on different training conditions to study some specific requirements such as the number of training instances [32].…”

Section: Introductionmentioning

confidence: 99%

“…As mentioned above, this is explained by the difficulty of having a clean real dataset or a real dataset where feature and label noise is clearly identified [40]. To overcome such limitations, some studies have analyzed class label noise influence firstly on synthetic data, and then on real data [38,41,42]. Experiments with synthetic data are needed since noise level is completely under control; whereas experiments with real datasets better represent data complexity, but are not guaranteed label noise-free.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Effect of Training Class Label Noise on Classification Performances for Land Cover Mapping with Satellite Image Time Series

et al. 2017

View full text Add to dashboard Cite

Supervised classification systems used for land cover mapping require accurate reference databases. These reference data come generally from different sources such as field measurements, thematic maps, or aerial photographs. Due to misregistration, update delay, or land cover complexity, they may contain class label noise, i.e., a wrong label assignment. This study aims at evaluating the impact of mislabeled training data on classification performances for land cover mapping. Particularly, it addresses the random and systematic label noise problem for the classification of high resolution satellite image time series. Experiments are carried out on synthetic and real datasets with two traditional classifiers: Support Vector Machines (SVM) and Random Forests (RF). A synthetic dataset has been designed for this study, simulating vegetation profiles over one year. The real dataset is composed of Landsat-8 and SPOT-4 images acquired during one year in the south of France. The results show that both classifiers are little influenced for low random noise levels up to 25%-30%, but their performances drop down for higher noise levels. Different classification configurations are tested by increasing the number of classes, using different input feature vectors, and changing the number of training instances. Algorithm complexities are also analyzed. The RF classifier achieves high robustness to random and systematic label noise for all the tested configurations; whereas the SVM classifier is more sensitive to the kernel choice and to the input feature vectors. Finally, this work reveals that the cross-validation procedure is impacted by the presence of class label noise.

show abstract

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Effect of Training Class Label Noise on Classification Performances for Land Cover Mapping with Satellite Image Time Series

et al. 2017

View full text Add to dashboard Cite

show abstract

“…Existing literature either deals with the problem from a theoretical perspective (Natarajan et al, 2013), or focuses on the relation between label noise and classification complexity using some benchmark dataset (Garcia et al, 2015). In the remote sensing community, two closely re-lated works are (Goldblatt et al, 2016) and (Pelletier et al, 2017).…”

Section: Introductionmentioning

confidence: 99%

Effect of the Training Set Configuration on Sentinel-2-Based Urban Local Climate Zone Classification

Qiu

Schmitt

Ghamisi

et al. 2018

Int. Arch. Photogramm. Remote Sens. Spatial Inf. Sci.

View full text Add to dashboard Cite

ABSTRACT:As any supervised classification procedure, also Local Climate Zone (LCZ) mapping requires reliable reference data. These are usually created manually and inevitably include label noise, caused by the complexity of the LCZ class scheme as well as variations in cultural and physical environmental factors. This study aims at evaluating the impact of the training set configuration, i.e. training sample number and imbalance, on the performance of Canonical Correlation Forests (CCFs) for a classification of the 11 urban LCZ classes. Experiments are carried out based on globally available Sentinel-2 imagery. Besides multi-spectral observations, different index measures extracted from the images as well as the Global Urban Footprint (GUF) and Open Street Map (OSM) layers are fed into the CCFs classifier. The results show that different LCZs favor different configurations in terms of training sample number and balance. Based on the findings, majority voting of different predictions from different configurations is proposed and performed. This way, a significant accuracy improvement can be achieved.

show abstract

“…Other work try to obtain noisetolerant classifiers by learning a label noise model jointly to the classification model [9]. In this case some information must be available about the label noise or its effects [2], [10]. The learning algorithm can also be modified to embed a data cleansing step [11].…”

Section: Noise Filteringmentioning

confidence: 99%

Adapting Noise Filters for Ranking

Lorena

García

Carvalho

2015

2015 Brazilian Conference on Intelligent Systems (BRACIS)

View full text Add to dashboard Cite

Abstract-Noise filtering can be considered an important preprocessing step in the data mining process, making data more reliable for pattern extraction. An interesting aspect for increasing data understanding would be to rank the potential noisy cases, in order to evidence the most unreliable instances to be further examined. Since the majority of the filters from the literature were designed only for hard classification, distinguishing whether an example is noisy or not, in this paper we adapt the output of some state of the art noise filters for ranking the cases identified as suspicious. We also present new evaluation measures for the noise rankers designed, which take into account the ordering of the detected noisy cases.

show abstract

Effect of label noise in the complexity of classification problems

Cited by 120 publications

References 28 publications

Effect of Training Class Label Noise on Classification Performances for Land Cover Mapping with Satellite Image Time Series

Effect of Training Class Label Noise on Classification Performances for Land Cover Mapping with Satellite Image Time Series

Effect of the Training Set Configuration on Sentinel-2-Based Urban Local Climate Zone Classification

Adapting Noise Filters for Ranking

Contact Info

Product

Resources

About