Szczegóły publikacji
Opis bibliograficzny
Enhancing predictive capabilities with analogy associative graphs for incomplete datasets / Konrad Komnata, Jakub KOMNATA, Adrian HORZYK, Wojciech Komnata, Mateusz Szyd // W: Neural Information Processing : 31st International Conference, ICONIP 2024 : Auckland, New Zealand, December 2–6, 2024 : proceedings, Pt. 2 / eds. Mufti Mahmud, [et al.]. — Singapore : Springer, cop. 2026. — (Communications in Computer and Information Science ; ISSN 1865-0929 ; CCIS 2283). — ISBN: 978-981-96-6950-9; e-ISBN: 978-981-96-6951-6. — S. 328–343. — Bibliogr., Abstr. — Publikacja dostępna online od: 2025-07-07. — J. Komnata – dod. afiliacja: Nivalit, Krakow, Poland
Autorzy (5)
- Komnata Konrad
- AGHKomnata Jakub
- AGHHorzyk Adrian
- Komnata Wojciech
- Szyd Mateusz
Dane bibliometryczne
| ID BaDAP | 161470 |
|---|---|
| Data dodania do BaDAP | 2025-08-08 |
| DOI | 10.1007/978-981-96-6951-6_23 |
| Rok publikacji | 2026 |
| Typ publikacji | materiały konferencyjne (aut.) |
| Otwarty dostęp |  |
| Wydawca | Springer |
| Konferencja | International Conference on Neural Information Processing 2024 |
| Czasopismo/seria | Communications in Computer and Information Science |
Abstract
Managing missing data in classification tasks poses significant challenges due to the diversity and unpredictability of missingness mechanisms and types. Common methods, such as imputation or deletion, often fail to provide robust solutions across domains, leading to potentially deceptive conclusions due to data distortion. A detailed analysis of missing data types and mechanisms is crucial for selecting appropriate deletion or imputation methods, including single imputation, model-based approaches, or interpolation techniques. This paper introduces Analogy Associative Graphs, which were designed to enhance predictive accuracy for datasets with many incomplete entities, and an algorithm for making predictions. The proposed model leverages specific graph structures to infer missing features using analogies between similar entities. We evaluated this model against well-established machine learning models and popular methods, including the Support Vector Classifier (SVC), AutoSklearn Classifier (AutoML), and Multi-Layer Perceptron Neural Network (MLP) across multiple datasets. Our research results demonstrate high predictive accuracy, F1 score, and stability of our model, even with up to 40% missing data. This performance surpasses common models, which can struggle with incomplete data and often rely on extensive data preprocessing. The ability of our model to handle incomplete data without compromising data integrity or corrupting information establishes it as a versatile and reliable tool for classification tasks in sectors plagued by data incompleteness.
