Meta-GA: Feature Selection for Abnormal Blockchain Transactions Based on Meta-Learning and Genetic Algorithms

With the rapid development of blockchain network and the frequent occurrence of abnormal transactions, the feature selection of abnormal blockchain transactions has become a complex and challenging task. Aiming at the problems of low efficiency of manual analysis, low accuracy of high-dimensional data recognition and large amount of calculation in the existing methods, a Meta-GA feature selection model is proposed. The model consists of three main modules: a heuristic guidance module which provides direction search strategy; The dynamic evaluation and correction module uses meta learning method to quickly evaluate the performance of feature subset; The feature subset evolution module uses genetic algorithm to achieve global search in high-dimensional space. Compared with the original data set, the accuracy of feature subset selection is 97.87 %, F1-score is 97.43 %, and AUC is improved by 4.4 %. Training time reduced by 30%.