As the fault-based attacks are becoming a more pertinent threat in today's era of edge computing/internet-of-things, there is a need to streamline the existing tools for better accuracy and ease of use, so that we can gauge the attacker's power and a proper countermeasure can be devised in the long run. In this regard, we propose a machine learning (ML) assisted tool that can be used in the context of a differential fault attack. In particular, finding the exact fault location by analysing the output difference (typically the XOR of the nonfaulty and the faulty ciphertexts) is somewhat nontrivial. During the literature survey, we notice that the Pearson's correlation coefficient dominantly is used for this purpose, and has almost become the defacto standard. While this method can yield good accuracy for certain cases, we argue that an ML-based method is more powerful in all the situations we experiment with. We substantiate our claim by showing the relative performances (we choose the commonly used multilayer perceptron as our ML tool) with two variants of Grain-128a (a stream cipher, and a stream cipher with authentication), the lightweight stream cipher LIZARD and the lightweight block cipher SIMON-32 (where the faults are injected at the fifth last rounds). Our results demonstrate that a common ML tool can outperform the correlation with the same training/testing data. We believe that our work extends the state-of-the-art by showing how traditional cryptographic methods can be replaced by a more powerful ML tool. © 2021 The Authors. CAAI Transactions on Intelligence Technology published by John Wiley & Sons Ltd on behalf of The Institution of Engineering and Technology and Chongqing University of Technology.