Completing right-censored data in time-series modelling

This paper focuses on nonparametric regression modelling of time series observations with data irregularities, such as censoring due to a cutoff value. In general, researchers do not prefer to put up with censored cases in time series analyses because their results are generally biased. In this paper, we present two imputation algorithms for handling auto-correlated censored data: artificial neural network based imputation and Gaussian imputation. These algorithms provide an estimation of the censored data points and replace them with their estimates. After the imputation procedure, the right-censored time series data is modelled and the performance of imputation methods are monitored. Thus, the effect of two imputation methods is evaluated for both modelling and completing censored observations. The purpose of this study is to prepare the censored data set for analysis without manipulating the observed part of the data such as synthetic data transformation or Kaplan-Meier weights. In this paper, algorithms for two methods are given and imputation processes are expressed.