Quantitative Performance Benchmarks for eDiscovery and Digital Forensics

This research paper presents quantitative performance benchmarks for two primary methodologies used in eDiscovery and digital forensics: keyword-based search and Technology-Assisted Review (TAR). The document synthesizes data from peer-reviewed studies and standardized evaluations, outlining the performance metrics of these methodologies. It details the typical ranges of recall, precision, and F1 scores for keyword searches, indicating that these methods often fall short of practitioner expectations in recall. The paper also discusses TAR performance, highlighting that TAR 2.0 (CAL) achieves high recall and precision, outperforming traditional human linear review methods. Additionally, it addresses limitations in evaluation, such as inter-assessor agreement and the impact of dataset prevalence on precision. The findings aim to provide defensible quantitative baselines for comparative evaluations of document classification systems, contributing to the understanding of performance in these critical areas of digital investigation.