Deep Forest and Pruned Syntax Tree-Based Classification Method for Java Code Vulnerability

The rapid development of J2EE (Java 2 Platform Enterprise Edition) has brought unprecedented severe challenges to vulnerability mining. current abstract syntax tree-based source code classification method does not eliminate irrelevant nodes when processing the tree, resulting in a long training time and overfitting problems. Another problem is that different structures will be translated same sequence tree trees using depth-first traversal, so this process, algorithm lead loss semantic structure information which reduce accuracy model. Aiming at these two problems, we propose deep forest pruned (PSTDF) for Java vulnerability. First, breadth-first traversal obtains statement trees, next, pruning removes nodes, then use based encoder obtain vector, finally, as classifier get results. Experiments on publicly accessible datasets show PSTDF can effectively remove impact redundant information.