A Unified Feedback-Driven Data Preparation Framework Using Persistent Feedback Artefacts

Abstract

Data preparation often requires analysts to make corrections that depend on the meaning and intended use of the data, not only its format. In many preparation tools, those corrections remain in the current file, script, or processing session and are unavailable when a similar problem appears in a later dataset. This study develops a Unified Feedback-Driven Data Preparation framework that incorporates the Persistent Feedback Artefact Model (PeFAM) into an end-to-end preparation workflow. The framework links dataset ingestion, staging, validation, transformation, feedback capture, persistent storage, analyst review, selective reuse, and export. A web prototype was implemented in PHP within a local XAMPP environment. CSV files were used for input and output, while MySQL records stored dataset settings, current-session feedback, and persistent feedback artefacts. The prototype retained validated corrections together with their original values and available contextual information. During a later task, it retrieved an artefact when the stored original value appeared in the current dataset and the affected attribute matched by exact column name or configured semantic role. The analyst then decided whether to select the suggestion. Selected items were copied to the active feedback store for use during transformation, while unselected items were not applied. The demonstration shows that these functions can operate within one preparation environment. The study does not evaluate efficiency, accuracy, scalability, or usability; those outcomes require a separate controlled evaluation. Its contribution is the operational connection between persistent artefact management and the core stages of data preparation.