Research On Classification Method For Mixed Frequency Data Based On Discriminative Dictionary Learning

A dilemma faced by classification is that data are not all collected at the same frequency in some real applications.In the scenario with mixed frequency data,the different features of data are sampled at different frequencies,and the numbers of corresponding samples are also different from each other.The traditional classification models will fail in such cases.We find the fact that mixed frequency data is a special style of multi-view data in which each view data is collected at different sampling frequencies.Thus,this thesis conducts the research on classification method for mixed frequency data based on discriminative dictionary learning.Aiming at solving the classification problems of mixed frequency data,this thesis proposes a classification method for mixed frequency data based on dictionary learning.Inspired on some classification ideas for multi-view data,the proposed model compares multiple data collected at multiple sampling frequencies to multiple views of multi-view data,and designs a way to learn a sub-dictionary for each class of each view.A structured dictionary whose dictionary atoms have correspondence to the class labels is learned.In addition,the Fisher discriminant criterion is used to constrain the within-class scatter and between-class scatter between the vectors of the coding coefficient matrix,thereby improving its discriminability.Finally,this thesis classifies samples by encoding and reconstructing them corresponding to the sampling frequency in the classification phase.Aiming at improving the discriminative ability of the learning dictionary and the performance of the model,this thesis improved the above method and proposed a novel discriminative dictionary pair learning constrained by ordinal locality for mixed frequency data classification.This method combines synthesis dictionary and analysis dictionary into a dictionary pair,which not only improves the training and test costs caused by traditional ₀L or ₁L norm constraint,but also can cope with inconsistent sampling frequency of mixed frequency data because of the class-specified sub-dictionary pair.The model utilizes synthesis dictionary to learn class-specified reconstruction information and utilizes analysis dictionary to generate coding coefficients by analyzing samples.Particularly,the ordinal locality preserving term is introduced to constrain the atoms of dictionaries pair to further facilitate the learned dictionary pair to be more discriminative.In this thesis,the eight real datasets,such as Sens IT,Digit,Web KB,are used to validate the effectiveness of the proposed method.The experimental results show that,in most cases,the proposed method is comparable to the best performing comparison method,or better than that method by 1 to 2 percentage points on the indicators of ACC and F1-score.It further illustrates that it is not advisable to fill mixed frequency data in some scenarios,thereby verifying the effectiveness of the classification method for mixed frequency data.