Multi-patterns discovering for time series based on kernel density estimation
Xiaoyun Chen · Journal of Shandong University · 2011
Existing discovery algorithms of frequent patterns or anomaly patterns can only find one of the two patterns from time series data,and most of them use the hard-distance threshold strategy.The FAP algorithm for mining frequent patterns and anomaly patterns at the same time was presented.The FAP algorithm adopted Gaussian kernel density as the support measure of the pattern,which uses a minimum density entropy approach to select a bandwidth parameter of the Gaussian kernel density function and avoids the defect of the hard-distance threshold strategy,so as to achieve multi-patterns discovering for time series.Because real time series datasets are often large or have noise,Haar wavelet transform was applied to compress the original time series and filter the noise.The FAP algorithm was tested on the time series from UCR data set,and the experimental results showed that the frequent patterns and anomaly patterns could be correctly found from these time series.