@unpublished{
 title = {Methods and architectures for risk and time-to-event modelling},
 type = {unpublished},
 year = {2018},
 institution = {University College London},
 id = {d8f5faf1-c90e-3f04-a9db-6bdb90e0430b},
 created = {2019-01-11T17:09:36.804Z},
 file_attached = {false},
 profile_id = {ff365b13-c449-37cb-8070-a8aedd49fa3c},
 last_modified = {2020-04-11T11:06:11.198Z},
 read = {true},
 starred = {false},
 authored = {true},
 confirmed = {true},
 hidden = {false},
 citation_key = {Sonabend2018},
 private_publication = {false},
 abstract = {In this thesis we present an overview of methods and architecture in risk and time-to-event modelling and introduce a novel reduction approach to survival modelling. By using discretisation to transform time-series into discrete panel data we are able to extend any standard o -shelf machine learning algorithm for a survival task. Via binning we show a simple method for creating conditional histogram estimates for a hazard function in survival analysis. To assess models against more traditional ones that only predict one time-point we de ne an aggregation method for averaging measures over correlated time-points in the presence of censoring. We demonstrate three key algorithms for extending o -shelf methods which can extend any classical machine learning model and propose meta-strategies for these algorithms. In the presence of survival data, we propose methods to deal with left and right censoring, including demonstrating possible biases that occur by ignoring right-censored data.},
 bibtype = {unpublished},
 author = {Sonabend, Raphael and Király, Franz J.}
}

{"_id":"vv6GyLxJMNJizakZW","bibbaseid":"sonabend-kirly-methodsandarchitecturesforriskandtimetoeventmodelling-2018","authorIDs":[],"author_short":["Sonabend, R.","Király, F., J."],"bibdata":{"title":"Methods and architectures for risk and time-to-event modelling","type":"unpublished","year":"2018","institution":"University College London","id":"d8f5faf1-c90e-3f04-a9db-6bdb90e0430b","created":"2019-01-11T17:09:36.804Z","file_attached":false,"profile_id":"ff365b13-c449-37cb-8070-a8aedd49fa3c","last_modified":"2020-04-11T11:06:11.198Z","read":"true","starred":false,"authored":"true","confirmed":"true","hidden":false,"citation_key":"Sonabend2018","private_publication":false,"abstract":"In this thesis we present an overview of methods and architecture in risk and time-to-event modelling and introduce a novel reduction approach to survival modelling. By using discretisation to transform time-series into discrete panel data we are able to extend any standard o -shelf machine learning algorithm for a survival task. Via binning we show a simple method for creating conditional histogram estimates for a hazard function in survival analysis. To assess models against more traditional ones that only predict one time-point we de ne an aggregation method for averaging measures over correlated time-points in the presence of censoring. We demonstrate three key algorithms for extending o -shelf methods which can extend any classical machine learning model and propose meta-strategies for these algorithms. In the presence of survival data, we propose methods to deal with left and right censoring, including demonstrating possible biases that occur by ignoring right-censored data.","bibtype":"unpublished","author":"Sonabend, Raphael and Király, Franz J.","bibtex":"@unpublished{\n title = {Methods and architectures for risk and time-to-event modelling},\n type = {unpublished},\n year = {2018},\n institution = {University College London},\n id = {d8f5faf1-c90e-3f04-a9db-6bdb90e0430b},\n created = {2019-01-11T17:09:36.804Z},\n file_attached = {false},\n profile_id = {ff365b13-c449-37cb-8070-a8aedd49fa3c},\n last_modified = {2020-04-11T11:06:11.198Z},\n read = {true},\n starred = {false},\n authored = {true},\n confirmed = {true},\n hidden = {false},\n citation_key = {Sonabend2018},\n private_publication = {false},\n abstract = {In this thesis we present an overview of methods and architecture in risk and time-to-event modelling and introduce a novel reduction approach to survival modelling. By using discretisation to transform time-series into discrete panel data we are able to extend any standard o -shelf machine learning algorithm for a survival task. Via binning we show a simple method for creating conditional histogram estimates for a hazard function in survival analysis. To assess models against more traditional ones that only predict one time-point we de ne an aggregation method for averaging measures over correlated time-points in the presence of censoring. We demonstrate three key algorithms for extending o -shelf methods which can extend any classical machine learning model and propose meta-strategies for these algorithms. In the presence of survival data, we propose methods to deal with left and right censoring, including demonstrating possible biases that occur by ignoring right-censored data.},\n bibtype = {unpublished},\n author = {Sonabend, Raphael and Király, Franz J.}\n}","author_short":["Sonabend, R.","Király, F., J."],"bibbaseid":"sonabend-kirly-methodsandarchitecturesforriskandtimetoeventmodelling-2018","role":"author","urls":{},"downloads":0},"bibtype":"unpublished","creationDate":"2019-12-19T10:02:25.374Z","downloads":0,"keywords":[],"search_terms":["methods","architectures","risk","time","event","modelling","sonabend","király"],"title":"Methods and architectures for risk and time-to-event modelling","year":2018}