Power of Fuzzing and Machine Learning in Smart Contract Security Validation

Abstract

Blockchain technology revolutionized digital payments and transactions by introducing disruptive capabilities. It operates through smart contracts, automated software code facilitating transactions without intermediaries. Smart contracts, written in various languages, have gained popularity but are vulnerable to logic flaws and security threats, potentially leading to financial losses and undermining blockchain integrity. Validating their security is essential, although current tools only detect specific attacks, lacking a comprehensive automated solution. This article presents a two-stage process wherein the initial phase involves the extraction of characteristics from smart contracts through the analysis of abstract syntax trees (ASTs) and control flow graphs (CFGs). In order to improve the precision of our results, we can leverage the distinctive capabilities of the fuzzing technique to label the data prior to its integration into the training model. In the subsequent phase, approach is utilized neural decision tree (NDT) typically combines a decision tree structure with neural network components. The mixed-method approach is utilized, leveraging the distinct advantages offered by neural networks and decision trees. The purpose of this collaborative technique is to enhance the probability of identifying vulnerabilities in smart contracts. The experimental assessment yielded favorable outcomes in the detection of smart contract vulnerabilities, namely those pertaining to Reentrancy, integer overflow, and Block Number Dependency.