Abstract
This research outlines how tax administrations can improve audit efficiency even when data quality, integration, and analytical staffing remain constrained.
This paper introduces a practical machine learning framework for identifying higher-risk taxpayers in environments where structured data is incomplete and systems are fragmented.
The core argument is that better audit targeting does not require perfect data. It requires disciplined feature selection, institutional realism, and workflows that tax authorities can actually operate.
The framework focuses on filing patterns, financial ratios, sectoral benchmarks, and behavior-aware risk scoring. It is designed to support low-capacity administrations that need measurable gains in compliance and enforcement outcomes.