SIFT: Sifting file types—application of explainable artificial intelligence in cyber forensics

Artificial Intelligence (AI) is being applied to improve the efficiency of software systems used in various domains, especially in the health and forensic sciences. Explainable AI (XAI) is one of the fields of AI that interprets and explains the methods used in AI. One of the techniques used in XAI to provide such interpretations is by computing the relevance of the input features to the output of an AI model. File fragment classification is one of the vital issues of file carving in Cyber Forensics (CF) and becomes challenging when the filesystem metadata is missing. Other major challenges it faces are: proliferation of file formats, file embeddings, automation, We leverage and utilize interpretations provided by XAI to optimize the classification of file fragments and propose a novel sifting approach, named SIFT (Sifting File Types). SIFT employs TF-IDF to assign weight to a byte (feature), which is used to select features from a file fragment. Threshold-based LIME and SHAP (the two XAI techniques) feature relevance values are computed for the selected features to optimize file fragment classification. To improve multinomial classification, a Multilayer Perceptron model is developed and optimized with five hidden layers, each layer with \(i \times n\) neurons, where i = the layer number and n = the total number of classes in the dataset. When tested with 47,482 samples of 20 file types (classes), SIFT achieves a detection rate of 82.1% and outperforms the other state-of-the-art techniques by at least 10%. To the best of our knowledge, this is the first effort of applying XAI in CF for optimizing file fragment classification.