Predictions Based on the Rectification and Processing of Noisy Data

Abstract

The described methods are devoted to developing of some strategies to avoid unnecessary hospital admissions on the base of a prediction of the possible future days in hospital based on previous claims and previous days in hospital statistics of a patient. Since often input data are not complete, a method of missing data restoring on the base of a similarity principle was represented first. The wishful prediction method is elaborated after data clustering and establishing of relations between different clusters. Also another approach to the prediction was introduced: the idea is based on a modeling of the Markov process with relation to the data. Both methods facilitate the production of algorithms and further precise calculations. More general computational topology method for different type of predictions was elaborated as well.