Deep learning-based sequential pattern mining for progressive database

摘要

Sequential pattern mining (SPM) is one of the main application areas in the field of online business, e-commerce, bioinformatics, etc. The traditional approaches in SPM are unable to accurately mine the huge volume of data. Therefore, the proposed work employs a sequential mining model based on deep learning to minimize complexity in handling huge data. Application areas such as online retailing, finance, and e-commerce face a dynamic change in data, which results in non-stationary data. Therefore, our proposed work uses discrete wavelet analysis to convert non-stationary data into time series. In the proposed SPM, a reformed hybrid combination of convolutional neural network (CNN) with long short-term memory (LSTM) is designed to find out customer behavior and purchasing patterns in terms of time. CNN is used to find the concerned itemsets (frequent) at the end of the pattern and LSTM for finding the time interval among each pair of successive itemsets. The proposed work mines the sequential pattern from a progressive database that removes the obsolete data. Finally, the accuracy of the proposed work is compared with some traditional algorithms to demonstrate its robustness.