Recognition of incomplete sequences using Fisher scores and hidden Markov models
Vadim E. Uvarov, Alexander A. Popov, Tatyana A. Gultyaeva · Journal of Physics Conference Series · 2018
We propose a method for recognition of incomplete sequences which makes use of feature vectors built from fisher scores for log likelihood function of hidden Markov model. We use support vector machines method to classify the feature vectors. The proposed method was compared to a previously developed method for incomplete sequence recognition based on marginalization of missing observations which can be considered as a state of the art method. The proposed method proved to be more effective than the SOTA method in situations when the percent of missing observations in training and testing sequences is high (more than 20% in our experiment). Thus, we suggest using the proposed method in situations when big percent of data is missing but the recognition still must be done.