Conditional Hazard Estimating Neural Networks

摘要: Survival analysis is used when we wish to study the occurrence of some event in a population subjects and time until interest. This called survival or failure time. often industrial life-testing experiments clinical follow-up studies. Examples application include: light bulb, an anomaly electronic circuit, relapse cancer, pregnancy. In literature find many different modeling approaches analysis. Conventional parametric models may involve too strict assumptions on distributions times form influence system features time, which usually extremely simplify experimental evidence, particularly case medical data (Cox & Oakes, 1984). contrast, semiparametric do not make failures, but instead how (the usual assumption proportionality hazards); furthermore, these allow for direct estimation times. Finally, non-parametric only qualitative description level. Neural networks have recently been analysis; survey current use neural networks, previous attempts at network refer (Bakker Heskes, 1999), (Biganzoli et al., 1998), (Eleuteri 2003), (Lisboa (Neal, 2001), (Ripley Ripley, (Schwarzer al. 2000). provide efficient estimates functions, and, principle, capability give personalised predictions. context, such information valuable both clinicians patients. It helps choose appropriate treatment plan efficiently. Patients high risk could be followed up more frequently than those lower order channel resources who need them most. For patients, obtaining about their prognosis also terms planning lives providing care dependents. this article describe novel model aimed solving problem continuous setting; details Bayesian approach modeling, sample real shown.