AI-Driven Forensic Methodology for Detecting and Classifying Factory Reset Timelines on Android Smartphones

View at publisher Summarise with VIRAT
Authors: A S Deepa, B. Bindhu, S P Sunil
Cited by: 0
References: 13

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Smartphones have become central to modern life, storing vast amounts of personal and transactional data and consequently serving as critical sources of evidence in digital forensic investigations. One of the most common anti-forensic tactics employed by suspects is the factory reset, intended to obscure incriminating data by restoring a device to its default state. However, residual artifacts such as system logs, configuration files, synchronization metadata and application preferences often persist post-reset, offering investigators opportunities to reconstruct timelines of user activity. Conventional forensic tools like UFED, Magnet AXIOM and MSAB XRY can extract these traces but require examiners to manually correlate fragmented data, a process that is time-consuming, error-prone and subject to interpretation bias. To address these limitations, this study introduces an AI-driven forensic methodology for detecting and classifying factory reset timelines on Android smartphones. Controlled experiments were conducted on a Xiaomi Poco F1 running Android 10, generating a dataset of system and application artifacts across varied reset conditions. The methodology integrates Random Forest classifiers for reset scenario classification, Gradient Boosted Regression Trees (GBRT) for timeline prediction and an Autoencoder for anomaly detection, combined with ensemble validation against outputs from professional forensic tools. Results demonstrate great performance: the Random Forest achieved$\mathbf{9 2 \%}$classification accuracy across four reset categories, with Immediate Setup and Initial Setup Screen scenarios yielding F1-scores of 0.93. The GBRT regression model achieved a mean absolute error (MAE) of 118 seconds, a root mean square error (RMSE) of 164 seconds and an$\mathbf{R}^{\mathbf{2}}$of 0.947, confirming precise reset time reconstruction. Anomaly detection achieved an AUC of 0.962 with a detection rate of 94.1 % and false positive rate of 5.8 %. Cross-validation showed AI predictions aligned within$\pm 2$seconds of ground-truth and forensic tool outputs. These findings underscore the potential of AI as a reliable enabler of forensic admissibility, offering examiners scalable, accurate and time-efficient solutions for reset detection and timeline reconstruction.

Fields of Study & Subjects

Digital and Cyber Forensics Digital Media Forensic Detection Advanced Malware Detection Techniques

Publication details

Journal / source
International Conference on Communication, Computing and Electronics Systems (ICCCES)
Pages
1267–1274
Publication date
21 January 2026
Language
EN