Year: 2019
Type: conference-paper
Abstract: In recent years, machine learning and deep learning have yielded immense success on computer vision, speech recognition and natural language processing. However, exploration of machine learning on software code weakness detect haven’t been fully investigated. Software source code could be thought of a kind of natural language but has the intrinsic difference from common natural language. In this work we propose a novel machine learning based approach to tackle the problem of software code weakness detection. Differing from traditional algorithm based tools, flawfinder, for example, we do not design the code weakness detection processing logics directly, instead we propose to use a variant version of Attribute-Relation File Format (ARFF) to preprocess the source code and utilize a state of the art classification algorithm in machine learning to build a unite code weakness detection model to learn the processing logic indirectly as a general framework. In order to exam the detection performance of our model, we collected a couple of datasets for each common weakness of Common Weakness Enumeration (CWE) List. Extensive results on these collected datasets show that our model achieves comparative results from traditional algorithm based tools in both false positive rate and false negative rate.
Authors: AiGuo Lu, KangYi Luo
Language: English
Subfield: Information Systems
Field: Computer Science
Domain: Physical Sciences
Open Access status: diamond
OpenAlex ID: W2949741054
