A Multi-Task Hybrid Deep Learning Based Framework for Robust File Fragment Classification in Digital Forensics

File fragment categorization is integral to the current digital forensics, as the failure to precisely retrieve and categorize fragmented data can compromise criminal investigations, cybersecurity incident responses, especially the reconstruction of vital evidence. Conventional machine learning techniques, including Support Vector Machines (SVM) as well as deep learning architectures that employ Long Short-Term Memory (LSTM), Convolutional Neural Network (CNN) networks, frequently struggle with the incomplete, high-dimensional characteristics of raw byte sequences and their vulnerability to noise. This research offers a Multi-Task Hybrid algorithm incorporating CNN, LSTM and Attention Layer framework that integrates generative modeling with multi-task learning to derive robust latent features while concurrently enhancing the reliability of classification. This hybrid model demonstrates superior performance compared to Random Forest, Fully Connected Convolutional Neural Network (FCCNN) and 1D Convolutional Neural Network (1D-CNN) baselines when assessed on a demanding dataset including 28 file types—5,505 fragments (4096 bytes each) derived from 16 subsets of the govdocs1 corpus in the Digital Corpora. The framework's capacity to unravel intricate byte-level patterns and reduce noise creates a novel norm, enhancing forensic skills in fragmented data recovery and setting a baseline for further studies in computational forensics.