Missing Attribute Clustering Eval

missing-attribute-clustering-eval

Clustering and Classification with Non-Existence Attributes: A Sentenced Discrepancy Measure Based Technique — Joarder et al. (2020) (arXiv:2002.10411, 2020)

What this evaluates

Tests clustering and classification algorithms on tabular datasets containing missing attributes (non-existence attributes). It evaluates how well direct discrepancy-based methods handle missing data under four simulated mechanisms (MCAR, MAR, MNAR-1, MNAR-2) compared to traditional imputation baselines.

Datasets

• Iris — total 150; splits: full (150)
• Sonar — total 208; splits: full (208)
• Glass — total 214; splits: full (214)
• Leaf — total 340; splits: full (340)
• Seeds — total 210; splits: full (210)
• Libras — total 360; splits: full (360)
• Chronic Kidney — total 800; splits: full (800)
• Vowel Context — total 990; splits: full (990)
• Isolate — total 1559; splits: full (1559)
• Landsat — total 6435; splits: full (6435)
• Breast Tissue — total 106; splits: full (106)
• Bank note — total 1372; splits: full (1372)

Metrics

• accuracy_rate (primary) — range: [0, 1]
  • Calculated as the proportion of correctly assigned cluster labels or class labels out of the total number of instances. No symbolic formula is provided in the paper; it follows standard classification/clustering accuracy conventions.

Input / output format

Input: Tabular feature vectors with some attributes explicitly removed (non-existence) according to MCAR, MAR, MNAR-1, or MNAR-2 mechanisms. All features are normalized to zero mean and unit standard deviation prior to evaluation.

Output: Predicted cluster labels or class labels for each instance.

Scoring recipe

def accuracy_rate(predictions, gold):
    correct = sum(1 for p, g in zip(predictions, gold) if p == g)
    return correct / len(gold)

Common pitfalls

• Missing data is artificially simulated under specific mechanisms rather than naturally occurring, which may limit real-world generalizability.
• All features are normalized to zero mean and unit standard deviation before evaluation, potentially obscuring scale-dependent algorithm behaviors.
• Results are reported as mean ± standard deviation, but the number of random seeds, initialization runs, or cross-validation folds is not explicitly stated.

Evidence (verbatim from paper)

The table and graphs represent the accuracy rate performances accuracy of Non-Existence attributes by MCAR Non-Existence using Scalable K-MEANS++-AWPD, K-MEANS++-AWPD, K-MEANS-FWPD and Imputation method with K-MEANS clustering algorithm.

Citation

@misc{joarder2020clustering,
  title={Clustering and Classification with Non-Existence Attributes: A Sentenced Discrepancy Measure Based Technique},
  author={Joarder et al. (2020)},
  year={2020},
  note={arXiv:2002.10411}
}

• arXiv: 2002.10411