There is a newer version of the record available.

Published November 4, 2026 | Version prereg-v1

missdetect: Hierarchical Classification of Missing-Data Mechanisms with Statistical and LLM-Augmented Features

Authors/Creators

  • 1. Instituto Tecnológico de Aeronáutica (ITA)

Description

missdetect is a research codebase that classifies the missingness mechanism (MCAR / MAR / MNAR) of a univariate column from a tabular dataset. It combines statistical and discriminative features with LLM-augmented features (judge, context-aware, self-consistency) and uses a two-level hierarchical classifier (Level 1: MCAR vs non-MCAR; Level 2: MAR vs MNAR) with Group-aware cross-validation and Cleanlab-weighted training for robustness to label noise.

Notes

If you use this software or the accompanying experiments, please cite both the software and the master's thesis.

Files

JonathanTsen/ITA-Mestrado-prereg-v1.zip

Files (78.6 MB)

Name Size
md5:10b713742ce0f88f8710f386164b987e
78.6 MB Preview Download

Additional details

Related works