br Fig Total possible enumeration for a
Fig. 5. Total possible enumeration for a discomfort value 10.
Fig. 6. Number of FA with sensitivity (in %) for various enumeration of discomfort.
, Probably Approximate Correct (PAC) learning of a binary classi-fier from noisy labeled examples acquired from multiple annotators , or a single unbiased estimator  may direct to a novel path for achieving the above objective. Also, chance-constrained programs , may be studied to design a conservative classifier which makes better tradeoffs between noise tolerance and making the FN zero.
Known artifacts before this study
• EMR helps in predicting cardiac problems, breast cancer, lung tumor and many other diseases but yet to contribute in predicting eso-phageal cancer, which is rising worldwide.
• Machine learning plays as a critical instrument in early detection of a disease.
Knowledge addition by this research
• Demographic, lifestyle and basic clinical data (without doctor's su-pervision) can predict Esophageal cancer with a very high accuracy.
• A feature transformation to a higher space can have a higher ac-curacy than orthodox machine learning methods and yield a high sensitivity to detect all true patients.
• Customized tests using a subset of features(tests) still can predict the esophageal cancer without compromising the probability of non inclusion of a true patient. • The novel idea of selecting a subset of standard tests to predict a disease can help many stakeholders – patient, doctor, insurance provider and others by reducing cost, improving quality or opti-mizing service parameters.
 Hoerres M, Critchley-Thorne R. Barrett's λ-Carrageenan and the need for improved di-agnostic and prognostic testing. 2014 [accessed 10.12.16]. http://www.mlo-online. com/barretts-esophagus-and-the-need-for-improved-diagnostic-and-prognostic-testing.php.
 Zhang H, Tan S, Chen W, Kligerman S, Kim G, D’Souza WD, et al. Modeling pa-thologic response of esophageal cancer to chemoradiation therapy using spatial-temporal fdg pet features, clinical parameters, and demographics. Int J Radiat Oncol Biol Phys 2014;88. 10.1016/j.ijrobp.2013.09.037.
 Macomber M, Samareh A, Chaovalitwongse W, Bowen S, Patel S, Zeng J, et al. Prediction of pathologic complete response to neoadjuvant chemoradiation in the treatment of esophageal cancer using machine learning. Int J Radiat Oncol Biol Phys 2016;96. 10.1016/j.ijrobp.2016.06.2379.
 Zieba M, Tomczak JM, Lubicz M, Światek J. Boosted SVM for extracting rules from imbalanced data in application to prediction of the post-operative life expectancy in the lung cancer patients. Appl Soft Comput 2014;14(Part A):99–108 special issue on hybrid intelligent methods for health technologies. 10.1016/j.asoc.2013.07.016.
 Scholkopf B, Sung K-K, Burges CJC, Girosi F, Niyogi P, Poggio T, et al. Comparing support vector machines with Gaussian kernels to radial basis function classifiers. Artificial Intelligence In Medicine 95 (2019) 16–26
 Fan R-E, Chang K-W, Hsieh C-J, Wang X-R, Lin C-J. LIBLINEAR: a periosteum library for large linear classification. J Mach Learn Res 2008;9:1871–4 Software available from: http://www.csie.ntu.edu.tw/cjlin/liblinear.  Schölkopf B, Smola AJ. Learning with kernels: support vector machines, regular-ization, optimization, and beyond. MIT Press; 2002.  Hall M, Frank E, Holmes G, Pfahringer B, Reutemann P, Witten IH. The weka data mining software: an update. SIGKDD Explorat 2009;11:1871–4 software available at: http://www.csie.ntu.edu.tw/cjlin/liblinear.  Chang C-C, Lin C-J. Libsvm: a library for support vector machines. ACM Trans Intell
26 International Journal of Radiation Oncology biology physics