168:
1991:"Imbalance correction led to models with strong miscalibration without better ability to distinguish between patients with and without the outcome event. The inaccurate probability estimates reduce the clinical utility of the model, because decisions about treatment are ill-informed.", The harm of class imbalance corrections for risk prediction models: illustration and simulation using logistic regression, 2022, Ruben van den Goorbergh, Maarten van Smeden, Dirk Timmerman, Ben Van Calster
66:
25:
895:. In this way, if two instances form a Tomek link then either one of these instances is noise or both are near a border. Thus, one can use Tomek links to clean up overlap between classes. By removing overlapping examples, one can establish well-defined clusters in the training set and lead to improved classification performance.
291:
such as years employed and current level of seniority. Suppose only 20% of software engineers are women, i.e., males are 4 times as frequent as females. If we were designing a survey to gather data, we would survey 4 times as many females as males, so that in the final sample, both genders will be represented equally. (See also
341:
Random
Oversampling involves supplementing the training data with multiple copies of some of the minority classes. Oversampling can be done more than once (2x, 3x, 5x, 10x, etc.) This is one of the earliest proposed methods, that is also proven to be robust. Instead of duplicating every sample in the
301:
Suppose we want to predict, from a large clinical dataset, which patients are likely to develop a particular disease (e.g., diabetes). Assume, however, that only 10% of patients go on to develop the disease. Suppose we have a large existing dataset. We can then pick 9 times the number of patients who
953:
Poor models in setting are often a result of—any combination of—fitting deterministic classifiers, using re-sampling or re-weighting methods to balance class frequencies in the training data and evaluating the model with a score such as accuracy. ... No re-sampling technique will magically generate
362:
features in the feature space of the data. Note that these features, for simplicity, are continuous. As an example, consider a dataset of birds for classification. The feature space for the minority class for which we want to oversample could be beak length, wingspan, and weight (all continuous). To
914:
Although sampling techniques have been developed mostly for classification tasks, growing attention is being paid to the problem of imbalanced regression. Adaptations of popular strategies are available, including undersampling, oversampling and SMOTE. Sampling techniques have also been explored in
306:
Oversampling is generally employed more frequently than undersampling, especially when the detailed data has yet to be collected by survey, interview or otherwise. Undersampling is employed much less frequently. Overabundance of already collected data became an issue only in the "Big Data" era, and
350:
There are a number of methods available to oversample a dataset used in a typical classification problem (using a classification algorithm to classify a set of images, given a labelled training set of images). The most common technique is known as SMOTE: Synthetic
Minority Over-sampling Technique.
290:
Suppose, to address the question of gender discrimination, we have survey data on salaries within a particular field, e.g., computer software. It is known women are under-represented considerably in a random sample of software engineers, which would be important when adjusting for other variables
386:
The adaptive synthetic sampling approach, or ADASYN algorithm, builds on the methodology of SMOTE, by shifting the importance of the classification boundary to those minority classes which are difficult. ADASYN uses a weighted distribution for different minority class examples according to their
323:
Data that is embedded in narrative text (e.g., interview transcripts) must be manually coded into discrete variables that a statistical or machine-learning package can deal with. The more the data, the more the coding effort. (Sometimes, the coding can be done through software, but somebody must
923:
It's possible to combine oversampling and undersampling techniques into a hybrid strategy. Common examples include SMOTE and Tomek links or SMOTE and Edited
Nearest Neighbors (ENN). Additional ways of learning on imbalanced datasets include weighing training instances, introducing different
1247:. This might be done in order to achieve "desireable", best performances for each class (potentially measured as precision and recall in each class). Finding the best multi-class classification performance or the best tradeoff between precision and recall is, however, inherently a
421:
Randomly remove samples from the majority class, with or without replacement. This is one of the earliest techniques used to alleviate imbalance in the dataset, however, it may increase the variance of the classifier and is very likely to discard useful or important samples.
315:
Domain experts will suggest dataset-specific means of validation involving not only intra-variable checks (permissible values, maximum and minimum possible valid values, etc.), but also inter-variable checks. For example, the individual components of a
282:
to select more samples from one class than from another, to compensate for an imbalance that is either already present in the data, or likely to develop if a purely random sample were taken. Data
Imbalance can be of the following types:
438:
Tomek links remove unwanted overlap between classes where majority class links are removed until all minimally distanced nearest neighbor pairs are of the same class. A Tomek link is defined as follows: given an instance pair
905:
A recent study shows that the combination of
Undersampling with ensemble learning can achieve better results, see IFME: information filtering by multiple examples with under-sampling in a digital library environment.
549:
395:
Data augmentation in data analysis are techniques used to increase the amount of data by adding slightly modified copies of already existing data or newly created synthetic data from existing data. It acts as a
940:
library. The re-sampling techniques are implemented in four different categories: undersampling the majority class, oversampling the minority class, combining over and under sampling, and ensembling sampling.
959:
Model
Comparison and Calibration Assessment User Guide for Consistent Scoring Functions in Machine Learning and Actuarial Practice, Tobias Fissler, arXiv:2202.12780v3, Christian Lorentzen, Michael Mayer,
83:
38:
265:
Oversampling and undersampling are opposite and roughly equivalent techniques. There are also more complex oversampling techniques, including the creation of artificial data points with algorithms like
311:
before it can be used. Cleansing typically involves a significant human component, and is typically specific to the dataset and the analytical problem, and therefore takes time and money. For example:
307:
the reasons to use undersampling are mainly practical and related to resource costs. Specifically, while one needs a suitably large sample size to draw valid statistical conclusions, the data must be
1058:
893:
809:
328:
For these reasons, one will typically cleanse only as much data as is needed to answer a question with reasonable statistical confidence (see Sample Size), but not more than that.
1220:
430:
Cluster centroids is a method that replaces cluster of samples by the cluster centroid of a K-means algorithm, where the number of clusters is set by the level of undersampling.
44:
598:
698:
483:
725:
652:
625:
1095:
1170:
1144:
1118:
130:
324:
often write a custom, one-off program to do so, and the program's output must be tested for accuracy, in terms of false positive and false negative results.)
1255:
solutions. Oversampling or undersampling as well as assigning weights to samples is an implicit way to find a certain pareto optimum (and it sacrifices the
102:
258:(i.e. the ratio between the different classes/categories represented). These terms are used both in statistical sampling, survey design methodology and in
109:
1311:
Kubat, M. (2000). Addressing the Curse of
Imbalanced Training Sets: One-Sided Selection. Fourteenth International Conference on Machine Learning.
116:
351:
However, this technique has been shown to yield poorly calibrated models, with an overestimated probability to belong to the minority class.
1837:
1613:
1334:
943:
The Python implementation of 85 minority oversampling techniques with model selection functions are available in the smote-variants package.
98:
488:
189:
1222:
useless and should be modified via undersampling or oversampling? The answer is no. Class imbalance is not a problem in itself at all.
1694:
229:
211:
149:
52:
915:
the context of numerical prediction in dependency-oriented data, such as time series forecasting and spatio-temporal forecasting.
1268:
1244:
267:
387:
level of difficulty in learning, where more synthetic data is generated for minority class examples that are harder to learn.
2055:
965:
123:
87:
1480:"The harm of class imbalance corrections for risk prediction models: illustration and simulation using logistic regression"
937:
1248:
397:
971:
814:
730:
182:
176:
1240:
76:
193:
1598:
2008 IEEE International Joint
Conference on Neural Networks (IEEE World Congress on Computational Intelligence)
1256:
1065:
933:
A variety of data re-sampling techniques are implemented in the imbalanced-learn package compatible with the
367:
nearest neighbors (in feature space). To create a synthetic data point, take the vector between one of those
1259:
of the estimated probabilities). A more explicit way than oversampling or downsampling could be to select a
1175:
375:
which lies between 0, and 1. Add this to the current data point to create the new, synthetic data point.
1288:
1479:
1275:
1814:. Lecture Notes in Computer Science. Vol. 8154. Berlin, Heidelberg: Springer. pp. 378–389.
1679:
IFME: information filtering by multiple examples with under-sampling in a digital library environment
317:
1590:
554:
1540:"SMOTE for Learning from Imbalanced Data: Progress and Challenges, Marking the 15-year Anniversary"
1267:
assign explicit costs to missclassified samples and then minimize the total (scalarized) costs via
657:
442:
292:
1278:
in a binary classification setting so that a certain validation precision and recall are achieved
2017:
1974:
1939:
1892:
1843:
1807:
1788:
1749:
1700:
1619:
1460:
1434:
1342:
Imbalanced-learn: A Python
Toolbox to Tackle the Curse of Imbalanced Datasets in Machine Learning
1478:
van den
Goorbergh, Ruben; van Smeden, Maarten; Timmerman, Dirk; Van Calster, Ben (2022-09-01).
1407:
Ling, Charles X., and Chenghui Li. "Data mining for direct marketing: Problems and solutions."
2004:
1931:
1884:
1833:
1741:
1690:
1609:
1571:
1517:
1499:
1452:
1330:
1293:
405:
378:
Many modifications and extensions have been made to the SMOTE method ever since its proposal.
1966:
1923:
1874:
1823:
1815:
1780:
1731:
1682:
1657:
1601:
1561:
1551:
1507:
1491:
1444:
1318:
259:
1365:
703:
630:
603:
2003:
Encyclopedia of Machine Learning. (2011). Deutschland: Springer. Page 193,
1538:
Chawla, Nitesh V.; Herrera, Francisco; Garcia, Salvador; Fernandez, Alberto (2018-04-20).
1071:
1768:
1539:
1422:
1149:
1123:
1341:
1512:
1388:
1260:
1252:
1103:
308:
2049:
1978:
1943:
1792:
1753:
1298:
1896:
1847:
1704:
1100:
This point can be illustrated with a simple example: Assume no predictive variables
2036:
Guillaume Lemaitre EuroSciPy 2023 - Get the best from your scikit-learn classifier
1623:
1464:
1251:
problem. It is well known that these problems typically have multiple incomparable
934:
1819:
1322:
1767:
Torgo, Luís; Branco, Paula; Ribeiro, Rita P.; Pfahringer, Bernhard (June 2015).
401:
65:
1927:
1910:
Oliveira, Mariana; Moniz, Nuno; Torgo, Luís; Santos Costa, Vítor (2021-09-01).
1736:
1719:
1605:
371:
neighbors, and the current data point. Multiply this vector by a random number
2037:
1879:
1862:
1662:
1645:
1061:
924:
misclassification costs for positive and negative examples and bootstrapping.
243:
1935:
1911:
1888:
1745:
1575:
1503:
1456:
354:
To illustrate how this technique works consider some training data which has
288:
Under-representation of a class in one or more important predictor variables.
1686:
1495:
1344:, Journal of Machine Learning Research, vol. 18, no. 17, 2017, pp. 1–5.
1315:
1806:
Torgo, Luís; Ribeiro, Rita P.; Pfahringer, Bernhard; Branco, Paula (2013).
1521:
1421:
Chawla, N. V.; Bowyer, K. W.; Hall, L. O.; Kegelmeyer, W. P. (2002-06-01).
254:
in data analysis are techniques used to adjust the class distribution of a
1970:
1912:"Biased resampling strategies for imbalanced spatio-temporal forecasting"
1556:
255:
2016:
Elor, Yotam; Averbuch-Elor, Hadar (2022). "To SMOTE, or not to SMOTE?".
1992:
1784:
1566:
1828:
1591:"ADASYN: Adaptive synthetic sampling approach for imbalanced learning"
342:
minority class, some of them may be randomly chosen with replacement.
299:
Under-representation of one class in the outcome (dependent) variable.
2022:
1448:
1393:
1370:
544:{\displaystyle x_{i}\in S_{\min },x_{j}\in S_{\operatorname {max} }}
302:
did not go on to develop the disease for every one patient who did.
1589:
He, Haibo; Bai, Yang; Garcia, Edwardo A.; Li, Shutao (June 2008).
1439:
363:
then oversample, take a sample from the dataset, and consider its
320:
must all add up to 100, because each is a percentage of the total.
1957:
Haibo He; Garcia, E.A. (2009). "Learning from Imbalanced Data".
279:
1863:"Resampling strategies for imbalanced time series forecasting"
161:
59:
18:
1810:. In Correia, Luís; Reis, Luís Paulo; Cascalho, José (eds.).
2005:
https://books.google.com/books?id=i8hQhp1a62UC&pg=PT193
1097:
during training by applying undersampling or downsampling.
954:
more information out of the few cases with the rare class.
278:
Both oversampling and undersampling involve introducing a
1677:
Zhu, Mingzhu; Xu, Chao; Wu, Yi-Fang Brook (2013-07-22).
1646:"A survey on Image Data Augmentation for Deep Learning"
1484:
Journal of the American Medical Informatics Association
1861:
Moniz, Nuno; Branco, Paula; Torgo, Luís (2017-05-01).
1178:
1152:
1126:
1106:
1074:
974:
817:
733:
706:
660:
633:
606:
557:
491:
445:
1720:"Imbalanced regression and extreme value prediction"
412:
Undersampling techniques for classification problems
1959:
IEEE Transactions on Knowledge and Data Engineering
1916:
International Journal of Data Science and Analytics
1867:
International Journal of Data Science and Analytics
1423:"SMOTE: Synthetic Minority Over-sampling Technique"
332:
Oversampling techniques for classification problems
90:. Unsourced material may be challenged and removed.
1360:
1358:
1214:
1164:
1138:
1112:
1089:
1052:
968:models trying to model a conditional distribution
887:
803:
719:
692:
646:
619:
592:
543:
477:
1053:{\displaystyle P(Y|X)={\frac {P(X|Y)P(Y)}{P(X)}}}
99:"Oversampling and undersampling in data analysis"
1644:Shorten, Connor; Khoshgoftaar, Taghi M. (2019).
1383:
1381:
1340:Lemaître, G. Nogueira, F. Aridas, Ch.K. (2017)
1316:Data Mining for Imbalanced Datasets: An Overview
888:{\displaystyle d(x_{j},x_{k})<d(x_{i},x_{j})}
804:{\displaystyle d(x_{i},x_{k})<d(x_{i},x_{j})}
510:
951:
700:is called a Tomek link if there's no instance
404:when training a machine learning model. (See:
274:Motivation for oversampling and undersampling
8:
1718:Ribeiro, Rita P.; Moniz, Nuno (2020-09-01).
1327:Data Mining and Knowledge Discovery Handbook
2038:https://www.youtube.com/watch?v=6YnhoCfArQo
1544:Journal of Artificial Intelligence Research
1427:Journal of Artificial Intelligence Research
53:Learn how and when to remove these messages
2021:
1878:
1827:
1735:
1661:
1565:
1555:
1511:
1438:
1180:
1179:
1177:
1151:
1125:
1105:
1073:
1010:
998:
984:
973:
876:
863:
841:
828:
816:
792:
779:
757:
744:
732:
711:
705:
681:
668:
659:
638:
632:
611:
605:
581:
568:
556:
535:
522:
509:
496:
490:
466:
453:
444:
268:Synthetic minority oversampling technique
230:Learn how and when to remove this message
212:Learn how and when to remove this message
150:Learn how and when to remove this message
175:This article includes a list of general
1650:Mathematics and Computers in Simulation
1366:"Scikit-learn-contrib/Imbalanced-learn"
1354:
1235:as well as assigning weights to samples
1769:"Resampling strategies for regression"
1068:if modifying the natural distribution
1993:https://doi.org/10.1093/jamia/ocac093
1325:In: Maimon, Oded; Rokach, Lior (Eds)
7:
1533:
1531:
1215:{\displaystyle {\hat {P}}(Y=1)=0.01}
900:Undersampling with ensemble learning
88:adding citations to reliable sources
1812:Progress in Artificial Intelligence
1389:"Analyticalmindsltd/Smote_variants"
1243:or situations with very imbalanced
1239:may be applied by practitioners in
318:differential white blood cell count
910:Techniques for regression problems
181:it lacks sufficient corresponding
14:
1172:is 0.99. Is a model which learns
34:This article has multiple issues.
166:
64:
23:
1269:cost-sensitive machine learning
75:needs additional citations for
42:or discuss these issues on the
16:Statistical sampling techniques
1203:
1191:
1185:
1146:is 0.01 and the proportion of
1084:
1078:
1044:
1038:
1030:
1024:
1018:
1011:
1004:
992:
985:
978:
966:Probabilistic machine learning
882:
856:
847:
821:
798:
772:
763:
737:
687:
661:
593:{\displaystyle d(x_{i},x_{j})}
587:
561:
472:
446:
1:
693:{\displaystyle (x_{i},x_{j})}
478:{\displaystyle (x_{i},x_{j})}
1820:10.1007/978-3-642-40669-0_33
1323:10.1007/978-0-387-09823-4_45
1249:multi-objective optimization
1120:and where the proportion of
2072:
1928:10.1007/s41060-021-00256-2
1737:10.1007/s10994-020-05900-9
1606:10.1109/IJCNN.2008.4633969
1241:multi-class classification
1880:10.1007/s41060-017-0044-3
1681:. ACM. pp. 107–110.
1663:10.1186/s40537-019-0197-0
1314:Chawla, Nitesh V. (2010)
600:is the distance between
1687:10.1145/2467696.2467736
196:more precise citations.
1808:"SMOTE for Regression"
1600:. pp. 1322–1328.
1301:(in signal processing)
1216:
1166:
1140:
1114:
1091:
1054:
963:
889:
805:
721:
694:
648:
621:
594:
545:
479:
2056:Sampling (statistics)
1971:10.1109/TKDE.2008.239
1496:10.1093/jamia/ocac093
1289:Sampling (statistics)
1217:
1167:
1141:
1115:
1092:
1055:
919:Additional techniques
890:
806:
722:
720:{\displaystyle x_{k}}
695:
649:
647:{\displaystyle x_{j}}
622:
620:{\displaystyle x_{i}}
595:
546:
480:
1557:10.1613/jair.1.11192
1176:
1150:
1124:
1104:
1090:{\displaystyle P(Y)}
1072:
972:
815:
731:
704:
658:
631:
604:
555:
489:
443:
417:Random undersampling
84:improve this article
1165:{\displaystyle Y=0}
1139:{\displaystyle Y=1}
337:Random oversampling
293:Stratified Sampling
1785:10.1111/exsy.12081
1397:. 26 October 2021.
1374:. 25 October 2021.
1212:
1162:
1136:
1110:
1087:
1064:) will be wrongly
1050:
885:
801:
717:
690:
644:
617:
590:
541:
475:
1839:978-3-642-40669-0
1615:978-1-4244-1820-6
1335:978-0-387-09823-4
1294:Data augmentation
1188:
1113:{\displaystyle X}
1048:
406:Data augmentation
400:and helps reduce
240:
239:
232:
222:
221:
214:
160:
159:
152:
134:
57:
2063:
2040:
2034:
2028:
2027:
2025:
2013:
2007:
2001:
1995:
1989:
1983:
1982:
1965:(9): 1263–1284.
1954:
1948:
1947:
1907:
1901:
1900:
1882:
1858:
1852:
1851:
1831:
1803:
1797:
1796:
1764:
1758:
1757:
1739:
1730:(9): 1803–1835.
1724:Machine Learning
1715:
1709:
1708:
1674:
1668:
1667:
1665:
1656:. springer: 60.
1641:
1635:
1634:
1632:
1630:
1595:
1586:
1580:
1579:
1569:
1559:
1535:
1526:
1525:
1515:
1490:(9): 1525–1534.
1475:
1469:
1468:
1449:10.1613/jair.953
1442:
1418:
1412:
1411:. Vol. 98. 1998.
1405:
1399:
1398:
1385:
1376:
1375:
1362:
1276:threshold tuning
1221:
1219:
1218:
1213:
1190:
1189:
1181:
1171:
1169:
1168:
1163:
1145:
1143:
1142:
1137:
1119:
1117:
1116:
1111:
1096:
1094:
1093:
1088:
1059:
1057:
1056:
1051:
1049:
1047:
1033:
1014:
999:
988:
961:
894:
892:
891:
886:
881:
880:
868:
867:
846:
845:
833:
832:
810:
808:
807:
802:
797:
796:
784:
783:
762:
761:
749:
748:
726:
724:
723:
718:
716:
715:
699:
697:
696:
691:
686:
685:
673:
672:
654:, then the pair
653:
651:
650:
645:
643:
642:
626:
624:
623:
618:
616:
615:
599:
597:
596:
591:
586:
585:
573:
572:
550:
548:
547:
542:
540:
539:
527:
526:
514:
513:
501:
500:
484:
482:
481:
476:
471:
470:
458:
457:
260:machine learning
235:
228:
217:
210:
206:
203:
197:
192:this article by
183:inline citations
170:
169:
162:
155:
148:
144:
141:
135:
133:
92:
68:
60:
49:
27:
26:
19:
2071:
2070:
2066:
2065:
2064:
2062:
2061:
2060:
2046:
2045:
2044:
2043:
2035:
2031:
2015:
2014:
2010:
2002:
1998:
1990:
1986:
1956:
1955:
1951:
1909:
1908:
1904:
1860:
1859:
1855:
1840:
1805:
1804:
1800:
1766:
1765:
1761:
1717:
1716:
1712:
1697:
1676:
1675:
1671:
1643:
1642:
1638:
1628:
1626:
1616:
1593:
1588:
1587:
1583:
1537:
1536:
1529:
1477:
1476:
1472:
1420:
1419:
1415:
1406:
1402:
1387:
1386:
1379:
1364:
1363:
1356:
1351:
1337:(pages 875–886)
1308:
1285:
1225:Additionally,
1174:
1173:
1148:
1147:
1122:
1121:
1102:
1101:
1070:
1069:
1034:
1000:
970:
969:
962:
958:
950:
930:
928:Implementations
921:
912:
901:
872:
859:
837:
824:
813:
812:
788:
775:
753:
740:
729:
728:
707:
702:
701:
677:
664:
656:
655:
634:
629:
628:
607:
602:
601:
577:
564:
553:
552:
531:
518:
505:
492:
487:
486:
462:
449:
441:
440:
436:
428:
419:
414:
393:
384:
348:
339:
334:
276:
236:
225:
224:
223:
218:
207:
201:
198:
188:Please help to
187:
171:
167:
156:
145:
139:
136:
93:
91:
81:
69:
28:
24:
17:
12:
11:
5:
2069:
2067:
2059:
2058:
2048:
2047:
2042:
2041:
2029:
2008:
1996:
1984:
1949:
1922:(3): 205–228.
1902:
1873:(3): 161–181.
1853:
1838:
1798:
1779:(3): 465–476.
1773:Expert Systems
1759:
1710:
1695:
1669:
1636:
1614:
1581:
1527:
1470:
1413:
1400:
1377:
1353:
1352:
1350:
1347:
1346:
1345:
1338:
1312:
1307:
1304:
1303:
1302:
1296:
1291:
1284:
1281:
1280:
1279:
1272:
1261:Pareto optimum
1253:Pareto optimal
1245:cost structure
1237:
1236:
1233:
1230:
1211:
1208:
1205:
1202:
1199:
1196:
1193:
1187:
1184:
1161:
1158:
1155:
1135:
1132:
1129:
1109:
1086:
1083:
1080:
1077:
1046:
1043:
1040:
1037:
1032:
1029:
1026:
1023:
1020:
1017:
1013:
1009:
1006:
1003:
997:
994:
991:
987:
983:
980:
977:
956:
949:
946:
945:
944:
941:
929:
926:
920:
917:
911:
908:
899:
884:
879:
875:
871:
866:
862:
858:
855:
852:
849:
844:
840:
836:
831:
827:
823:
820:
800:
795:
791:
787:
782:
778:
774:
771:
768:
765:
760:
756:
752:
747:
743:
739:
736:
714:
710:
689:
684:
680:
676:
671:
667:
663:
641:
637:
614:
610:
589:
584:
580:
576:
571:
567:
563:
560:
538:
534:
530:
525:
521:
517:
512:
508:
504:
499:
495:
474:
469:
465:
461:
456:
452:
448:
435:
432:
427:
424:
418:
415:
413:
410:
392:
389:
383:
380:
347:
344:
338:
335:
333:
330:
326:
325:
321:
304:
303:
296:
275:
272:
238:
237:
220:
219:
174:
172:
165:
158:
157:
72:
70:
63:
58:
32:
31:
29:
22:
15:
13:
10:
9:
6:
4:
3:
2:
2068:
2057:
2054:
2053:
2051:
2039:
2033:
2030:
2024:
2019:
2012:
2009:
2006:
2000:
1997:
1994:
1988:
1985:
1980:
1976:
1972:
1968:
1964:
1960:
1953:
1950:
1945:
1941:
1937:
1933:
1929:
1925:
1921:
1917:
1913:
1906:
1903:
1898:
1894:
1890:
1886:
1881:
1876:
1872:
1868:
1864:
1857:
1854:
1849:
1845:
1841:
1835:
1830:
1825:
1821:
1817:
1813:
1809:
1802:
1799:
1794:
1790:
1786:
1782:
1778:
1774:
1770:
1763:
1760:
1755:
1751:
1747:
1743:
1738:
1733:
1729:
1725:
1721:
1714:
1711:
1706:
1702:
1698:
1696:9781450320771
1692:
1688:
1684:
1680:
1673:
1670:
1664:
1659:
1655:
1651:
1647:
1640:
1637:
1625:
1621:
1617:
1611:
1607:
1603:
1599:
1592:
1585:
1582:
1577:
1573:
1568:
1563:
1558:
1553:
1549:
1545:
1541:
1534:
1532:
1528:
1523:
1519:
1514:
1509:
1505:
1501:
1497:
1493:
1489:
1485:
1481:
1474:
1471:
1466:
1462:
1458:
1454:
1450:
1446:
1441:
1436:
1432:
1428:
1424:
1417:
1414:
1410:
1404:
1401:
1396:
1395:
1390:
1384:
1382:
1378:
1373:
1372:
1367:
1361:
1359:
1355:
1348:
1343:
1339:
1336:
1332:
1328:
1324:
1320:
1317:
1313:
1310:
1309:
1305:
1300:
1299:Undersampling
1297:
1295:
1292:
1290:
1287:
1286:
1282:
1277:
1273:
1270:
1266:
1265:
1264:
1262:
1258:
1254:
1250:
1246:
1242:
1234:
1232:undersampling
1231:
1228:
1227:
1226:
1223:
1209:
1206:
1200:
1197:
1194:
1182:
1159:
1156:
1153:
1133:
1130:
1127:
1107:
1098:
1081:
1075:
1067:
1063:
1041:
1035:
1027:
1021:
1015:
1007:
1001:
995:
989:
981:
975:
967:
955:
947:
942:
939:
936:
932:
931:
927:
925:
918:
916:
909:
907:
903:
902:
896:
877:
873:
869:
864:
860:
853:
850:
842:
838:
834:
829:
825:
818:
793:
789:
785:
780:
776:
769:
766:
758:
754:
750:
745:
741:
734:
712:
708:
682:
678:
674:
669:
665:
639:
635:
612:
608:
582:
578:
574:
569:
565:
558:
536:
532:
528:
523:
519:
515:
506:
502:
497:
493:
467:
463:
459:
454:
450:
433:
431:
425:
423:
416:
411:
409:
407:
403:
399:
390:
388:
381:
379:
376:
374:
370:
366:
361:
358:samples, and
357:
352:
345:
343:
336:
331:
329:
322:
319:
314:
313:
312:
310:
300:
297:
294:
289:
286:
285:
284:
281:
273:
271:
269:
263:
261:
257:
253:
252:undersampling
249:
245:
234:
231:
216:
213:
205:
195:
191:
185:
184:
178:
173:
164:
163:
154:
151:
143:
132:
129:
125:
122:
118:
115:
111:
108:
104:
101: –
100:
96:
95:Find sources:
89:
85:
79:
78:
73:This article
71:
67:
62:
61:
56:
54:
47:
46:
41:
40:
35:
30:
21:
20:
2032:
2023:2201.08528v3
2011:
1999:
1987:
1962:
1958:
1952:
1919:
1915:
1905:
1870:
1866:
1856:
1811:
1801:
1776:
1772:
1762:
1727:
1723:
1713:
1678:
1672:
1653:
1649:
1639:
1627:. Retrieved
1597:
1584:
1547:
1543:
1487:
1483:
1473:
1430:
1426:
1416:
1408:
1403:
1392:
1369:
1326:
1238:
1229:oversampling
1224:
1099:
964:
952:
935:scikit-learn
922:
913:
904:
898:
897:
437:
429:
420:
394:
391:Augmentation
385:
377:
372:
368:
364:
359:
355:
353:
349:
340:
327:
305:
298:
287:
277:
264:
251:
248:oversampling
247:
241:
226:
208:
199:
180:
146:
137:
127:
120:
113:
106:
94:
82:Please help
77:verification
74:
50:
43:
37:
36:Please help
33:
1567:10481/56411
1550:: 863–905.
1433:: 321–357.
1329:, Springer
1257:calibration
434:Tomek links
402:overfitting
398:regularizer
194:introducing
1829:10289/8518
1629:5 December
1349:References
1306:Literature
1066:calibrated
1062:Bayes rule
727:such that
244:statistics
202:April 2011
177:references
140:April 2011
110:newspapers
39:improve it
1979:206742563
1944:210931099
1936:2364-4168
1889:2364-4168
1793:205129966
1754:222143074
1746:1573-0565
1576:1076-9757
1504:1527-974X
1457:1076-9757
1440:1106.1813
1186:^
1060:(through
948:Criticism
529:∈
503:∈
45:talk page
2050:Category
1897:25975914
1848:16253787
1705:13279787
1522:35686364
1283:See also
1274:perform
957:—
485:, where
256:data set
1624:1438164
1513:9382395
1465:1554582
426:Cluster
309:cleaned
242:Within
190:improve
124:scholar
1977:
1942:
1934:
1895:
1887:
1846:
1836:
1791:
1752:
1744:
1703:
1693:
1622:
1612:
1574:
1520:
1510:
1502:
1463:
1455:
1394:GitHub
1371:GitHub
1333:
938:Python
382:ADASYN
179:, but
126:
119:
112:
105:
97:
2018:arXiv
1975:S2CID
1940:S2CID
1893:S2CID
1844:S2CID
1789:S2CID
1750:S2CID
1701:S2CID
1620:S2CID
1594:(PDF)
1461:S2CID
1435:arXiv
346:SMOTE
131:JSTOR
117:books
1932:ISSN
1885:ISSN
1834:ISBN
1742:ISSN
1691:ISBN
1631:2022
1610:ISBN
1572:ISSN
1518:PMID
1500:ISSN
1453:ISSN
1331:ISBN
1210:0.01
960:2023
851:<
767:<
627:and
551:and
280:bias
250:and
103:news
1967:doi
1924:doi
1875:doi
1824:hdl
1816:doi
1781:doi
1732:doi
1728:109
1683:doi
1658:doi
1602:doi
1562:hdl
1552:doi
1508:PMC
1492:doi
1445:doi
1409:Kdd
1319:doi
1263:by
811:or
537:max
511:min
86:by
2052::
1973:.
1963:21
1961:.
1938:.
1930:.
1920:12
1918:.
1914:.
1891:.
1883:.
1869:.
1865:.
1842:.
1832:.
1822:.
1787:.
1777:32
1775:.
1771:.
1748:.
1740:.
1726:.
1722:.
1699:.
1689:.
1652:.
1648:.
1618:.
1608:.
1596:.
1570:.
1560:.
1548:61
1546:.
1542:.
1530:^
1516:.
1506:.
1498:.
1488:29
1486:.
1482:.
1459:.
1451:.
1443:.
1431:16
1429:.
1425:.
1391:.
1380:^
1368:.
1357:^
408:)
295:.)
270:.
262:.
246:,
48:.
2026:.
2020::
1981:.
1969::
1946:.
1926::
1899:.
1877::
1871:3
1850:.
1826::
1818::
1795:.
1783::
1756:.
1734::
1707:.
1685::
1666:.
1660::
1654:6
1633:.
1604::
1578:.
1564::
1554::
1524:.
1494::
1467:.
1447::
1437::
1321::
1271:.
1207:=
1204:)
1201:1
1198:=
1195:Y
1192:(
1183:P
1160:0
1157:=
1154:Y
1134:1
1131:=
1128:Y
1108:X
1085:)
1082:Y
1079:(
1076:P
1045:)
1042:X
1039:(
1036:P
1031:)
1028:Y
1025:(
1022:P
1019:)
1016:Y
1012:|
1008:X
1005:(
1002:P
996:=
993:)
990:X
986:|
982:Y
979:(
976:P
883:)
878:j
874:x
870:,
865:i
861:x
857:(
854:d
848:)
843:k
839:x
835:,
830:j
826:x
822:(
819:d
799:)
794:j
790:x
786:,
781:i
777:x
773:(
770:d
764:)
759:k
755:x
751:,
746:i
742:x
738:(
735:d
713:k
709:x
688:)
683:j
679:x
675:,
670:i
666:x
662:(
640:j
636:x
613:i
609:x
588:)
583:j
579:x
575:,
570:i
566:x
562:(
559:d
533:S
524:j
520:x
516:,
507:S
498:i
494:x
473:)
468:j
464:x
460:,
455:i
451:x
447:(
373:x
369:k
365:k
360:f
356:s
233:)
227:(
215:)
209:(
204:)
200:(
186:.
153:)
147:(
142:)
138:(
128:·
121:·
114:·
107:·
80:.
55:)
51:(
Text is available under the Creative Commons Attribution-ShareAlike License. Additional terms may apply.