Abstract
Some supervised tasks are presented with a numerical output but decisions have to be made in a discrete, binarised, way, according to a particular cutoff. This binarised regression task is a very common situation that requires its own analysis, different from regression and classification—and ordinal regression. We first investigate the application cases in terms of the information about the distribution and range of the cutoffs and distinguish six possible scenarios, some of which are more common than others. Next, we study two basic approaches: the retraining approach, which discretises the training set whenever the cutoff is available and learns a new classifier from it, and the reframing approach, which learns a regression model and sets the cutoff when this is available during deployment. In order to assess the binarised regression task, we introduce context plots featuring error against cutoff. Two special cases are of interest, the \( UCE \) and \( OCE \) curves, showing that the area under the former is the mean absolute error and the latter is a new metric that is in between a ranking measure and a residual-based measure. A comprehensive evaluation of the retraining and reframing approaches is performed using a repository of binarised regression problems created on purpose, concluding that no method is clearly better than the other, except when the size of the training data is small.
Similar content being viewed by others
Notes
Note that some people can buy a house that is much cheaper than its maximum mortgage, especially if they buy it as an investment or to refurbish it afterwards.
It is worth noting that the training process is entirely repeated in the retraining alternative, having nothing to do with any kind of incremental learning or adaptation of the previous model. This use of the term ‘retraining’, understood as building a different model each time a new cutoff is set, can often be found in the active learning research field (Guo and Schuurmans 2008; Sammut and Webb 2011).
Note that region is here used to refer to an interval (continuous subset of values) within all the possible cutoff values. This interval will usually be narrow.
For the interested reader, it is worth mentioning that Theorem 1 is connected to Theorem 11 (and corollary 12) by Hernández-Orallo et al. (2012), where the expected loss of the score-uniform threshold choice method for a uniform distribution of operating contexts (cost proportions or skews) is shown to be equal to \( MAE \). Two comments must be done, though. First, here we are talking about the \( MAE \) of a regression model while in Hernández-Orallo et al. (2012) the result holds for a soft classifier with estimated probabilities between 0 and 1—upon which the \( MAE \) is calculated. Second, here the decision rule is taking the operating context into account—the cutoff is used at each point of the curve, while in Hernández-Orallo et al. (2012) the result is obtained by the score-uniform threshold choice method, which completely ignores the operating context. Nevertheless, this is still an interesting connection as both are assuming a uniform distribution of contexts.
Quantile regression aims at estimating either the conditional median or other quantiles of the goal variable.
For example, consider three neighbours with outputs \(y= \{1, 2, 6\}\), and a cutoff \(c=2.5\). If we consider equal weights and mean for the prediction, for regression, the average \(\bar{y} = 4.5 \ge c = 2.5\) and predicts “above the cutoff”, but for classification there is only one neighbour above the cutoff so it predicts “below the cutoff”. Only for \(k=1\) would both approaches be equal.
References
Bache K, Lichman M (2013) UCI machine learning repository. http://archive.ics.uci.edu/ml
Bella A, Ferri C, Hernández-Orallo J, Ramírez-Quintana MJ (2014) Aggregative quantification for regression. Data Min Knowl Discov 28(2):475–518
Bi J, Bennett KP (2003) Regression error characteristic curves. In: Twentieth international conference on machine learning (ICML-2003). Washington, DC
Brooks AD (2007) knnflex: a more flexible KNN. R package version 1.1.1
Cohen I, Goldszmidt M (2004) Properties and benefits of calibrated classifiers. Knowl Discov Database 2004:125–136
Drummond C, Holte R (2000) Explicitly representing expected cost: an alternative to ROC representation. In: Knowledge discovery and data mining, pp 198–207
Drummond C, Holte R (2006) Cost curves: an improved method for visualizing classifier performance. Mach Learn 65:95–130
Fawcett T (2006) ROC graphs with instance-varying costs. Pattern Recognit Lett 27(8):882–891
Fawcett T, Provost F (1997) Adaptive fraud detection. Data Min Knowl Discov 1(3):291–316
Federal Financial Institutions Examination Council (2013) Home mortgage disclosure act (HMDA). http://www.ffiec.gov/hmda/
Ferri C, Hernández-Orallo J (2004) Cautious classifiers. In: Proceedings of the 1st international workshop on ROC analysis in artificial intelligence (ROCAI-2004), pp 27–36
Ferri C, Hernández-Orallo J, Modroiu R (2009) An experimental comparison of performance measures for classification. Pattern Recognit Lett 30(1):27–38
Flach P (2003) The geometry of ROC space: understanding machine learning metrics through ROC isometrics. In: Machine learning, proceedings of the twentieth international conference (ICML 2003), pp 194–201
Guo Y, Schuurmans D (2008) Discriminative batch mode active learning. In: Platt J, Koller D, Singer Y, Roweis S (eds) Advances in neural information processing systems, vol 20. Curran Associates, Inc, pp 593–600
Hall M, Frank E, Holmes G, Pfahringer B, Reutemann P, Witten IH (2009) The WEKA data mining software: an update. SIGKDD Explor Newsl 11(1):10–18
Hastie T, Tibshirani R, Friedman J (2001) The elements of statistical learning. Springer series in statistics. Springer New York Inc., New York
Hernández-Orallo J (2013) ROC curves for regression. Pattern Recognit 46(12):3395–3411
Hernández-Orallo J (2014) Probabilistic reframing for context-sensitive regression. ACM Trans Knowl Discov Data 8(3)
Hernández-Orallo J, Flach P, Ferri C (2012) A unified view of performance metrics: translating threshold choice into expected classification loss. J Mach Learn Res (JMLR) 13:2813–2869
Hornik K, Buchta C, Zeileis A (2009) Open-source machine learning: R meets Weka. Comput Stat 24(2):225–232. doi:10.1007/s00180-008-0119-7
Hsu CN, Knoblock CA (1998) Discovering robust knowledge from databases that change. Data Min Knowl Discov 2(1):69–95
Kocjan E, Kononenko I (2009) Regression as cost-sensitive classification. In: International multiconference on information society, pp 38–41
Koenker R (2005) Quantile regression, vol 38. Cambridge University Press, Cambridge
Langford J, Oliveira R, Zadrozny B (2012) Predicting conditional quantiles via reduction to classification. arXiv:1206.6860
Langford J, Zadrozny B (2005) Estimating class membership probabilities using classifier learners. In: Proceedings of the tenth international workshop on artificial intelligence and statistics (AISTAT05), pp 198–205
Martin A, Doddington G, Kamm T, Ordowski M, Przybocki M (1997) The DET curve in assessment of detection task performance. In: Fifth european conference on speech communication and technology. Citeseer
Pan SJ, Yang Q (2010) A survey on transfer learning. IEEE Trans Knowl Data Eng 22(10):1345–1359
Piatetsky-Shapiro G, Masand B (1999) Estimating campaign benefits and modeling lift. In: Proceedings of the fifth ACM SIGKDD international conference on Knowledge discovery and data mining. ACM, p 193
Pietraszek T (2007) On the use of ROC analysis for the optimization of abstaining classifiers. Mach Learn 68(2):137–169
Prati RC, Batista GE, Monard MC (2011) A survey on graphical methods for classification predictive performance evaluation. IEEE Trans Knowl Data Eng 23:1601–1618. doi:10.1109/TKDE.2011.59
Rosset S, Perlich C, Zadrozny B (2007) Ranking-based evaluation of regression models. Knowl Inf Syst 12(3):331–353
Sammut C, Webb G (2011) Encyclopedia of machine learning. Encyclopedia of machine learning. Springer, New York
Swets JA, Dawes RM, Monahan J (2000) Better decisions through science. Sci Am 283(4):82–87
Torgo L (2005) Regression error characteristic surfaces. In: Proceedings of the eleventh ACM SIGKDD international conference on Knowledge discovery in data mining. ACM, pp 697–702
Torgo L, Gama J (1996) Regression by classification. In: Advances in artificial intelligence. Springer, pp 51–60
The keel-dataset repository (2002). http://www.keel.es/
Yang Y, Wu X, Zhu X (2006) Mining in anticipation for concept change: proactive-reactive prediction in data streams. Data Min Knowl Discov 13(3):261–289
Zillow (2013) Zillow API. http://www.zillow.com/howto/api/APIOverview.htm
Acknowledgments
We thank the anonymous reviewers for their comments, which have helped to improve this paper significantly. We thank Peter Flach and Meelis Kull for their insightful comments and very useful suggestions. This work was supported by the Spanish MINECO under Grant TIN 2013-45732-C4-1-P and by Generalitat Valenciana PROMETEOII2015/013. This research has been developed within the REFRAME project, granted by the European Coordinated Research on Long-term Challenges in Information and Communication Sciences & Technologies ERA-Net (CHIST-ERA), and funded by the Ministerio de Economía y Competitividad in Spain (PCIN-2013-037) and the Agence Nationale pour la Recherche in France (ANR-12-CHRI-0005-03).
Author information
Authors and Affiliations
Corresponding author
Additional information
Responsible editor: Eamonn Keogh.
Rights and permissions
About this article
Cite this article
Hernández-Orallo, J., Ferri, C., Lachiche, N. et al. Binarised regression tasks: methods and evaluation metrics. Data Min Knowl Disc 30, 848–890 (2016). https://doi.org/10.1007/s10618-015-0443-9
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s10618-015-0443-9