Handling imbalance in hierarchical classification problems using local classifiers approaches

Pereira, Rodolfo Miranda; Costa, Yandre M. G.; Silla, Carlos N.

doi:10.1007/s10618-021-00762-8

Cited by 6 publications

(6 citation statements)

References 68 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…However, unlike our approach, where a single model is used to learn the overall problem, local classifiers require a more significant computation effort for learning separated multiple models [28], especially in situations where the complexity of the task increases. This evidence is also confirmed in our experiment as our HOBD and HCLM methods perform better than the state-of-the-art global and local alternatives [22], [25], [26], [27]. Another significant difference from all the above-cited works is how the proposed approach models the labels of each hierarchical level.…”

Section: B Hierarchical Classificationsupporting

confidence: 80%

“…However, a significant limitation of this approach is that it completely ignores the class relationships and any hierarchical constraints while typically predicting only the leaf nodes (flat classification). A different approach (local classifiers) [25], [26], [27] is to employ a set of classifiers for each node or each parent level. Thus, each classifier is specialized in solving the classification task associated with the child nodes.…”

Section: B Hierarchical Classificationmentioning

confidence: 99%

“…The ordinal variant of this approach was implemented by integrating ordinal relationship through OBD [5] (GLB-OBD) and CLM [39] (GLB-CLM). 2) Local Classifier per Parent Node (LCPN) [25], [26], [27] Approaches: Models are trained for solving each local task (i.e., a separate model for each parent node) using nominal (LCPN-NOM) or ordinal classifiers. The ordinal relationship was encoded with an OBD (LCPN-OBD) and a CLM (LCPN-CLM) layer by considering, respectively, MAE and QWK as loss functions.…”

Section: B Experimental Comparisonsmentioning

confidence: 99%

See 2 more Smart Citations

Learning Ordinal–Hierarchical Constraints for Deep Learning Classifiers

Rosati,

Romeo,

Vargas

et al. 2024

IEEE Trans. Neural Netw. Learning Syst.

View full text Add to dashboard Cite

Real-world classification problems may disclose different hierarchical levels where the categories are displayed in an ordinal structure. However, no specific deep learning (DL) models simultaneously learn hierarchical and ordinal constraints while improving generalization performance. To fill this gap, we propose the introduction of two novel ordinal-hierarchical DL methodologies, namely, the hierarchical cumulative link model (HCLM) and hierarchical-ordinal binary decomposition (HOBD), which are able to model the ordinal structure within different hierarchical levels of the labels. In particular, we decompose the hierarchical-ordinal problem into local and global graph paths that may encode an ordinal constraint for each hierarchical level. Thus, we frame this problem as simultaneously minimizing global and local losses. Furthermore, the ordinal constraints are set by two approaches [ordinal binary decomposition (OBD) and cumulative link model (CLM)] within each global and local function. The effectiveness of the proposed approach is measured on four real-use case datasets concerning industrial, biomedical, computer vision, and financial domains. The extracted results demonstrate a statistically significant improvement to state-ofthe-art nominal, ordinal, and hierarchical approaches.

show abstract

Section: B Hierarchical Classificationsupporting

confidence: 80%

Section: B Hierarchical Classificationmentioning

confidence: 99%

Section: B Experimental Comparisonsmentioning

confidence: 99%

See 1 more Smart Citation

Learning Ordinal–Hierarchical Constraints for Deep Learning Classifiers

Rosati,

Romeo,

Vargas

et al. 2024

IEEE Trans. Neural Netw. Learning Syst.

View full text Add to dashboard Cite

show abstract

“…Only a small number of studies try to address the imbalance problem for hierarchical classification. The proposed approaches, however, rely on local classifiers [S13] or transformation to label sets [S14] to perform this resampling. As explained earlier we chose to focus on global classifier approaches using binary relevance such that our results are most easily transferable to existing work on ICD classification.…”

Section: Appendixmentioning

confidence: 99%

Creating a computer assisted ICD coding system: performance metric choice and use of the ICD hierarchy

Marcou,

Berti-Equille,

Novelli

2024

Preprint

View full text Add to dashboard Cite

Objective: Machine learning methods hold the promise of leveraging available data and generating higher quality data while alleviating the data collection burden on healthcare professionals. International Classification of Diseases (ICD) diagnoses data, collected globally for billing and epidemiological purposes, represents a valuable source of structured information. However, ICD coding is a challenging task. While numerous previous studies reported promising results in automatic ICD classification, they often describe input data specific model architectures, that are heterogeneously evaluated with different performance metrics and ICD code subsets. This study aims to explore the evaluation and construction of more effective Computer Assisted Coding (CAC) systems using generic approaches, focusing on the use of ICD hierarchy, medication data and a feed forward neural network architecture. Methods: We conduct comprehensive experiments using the MIMIC-III clinical database, mapped to the OMOP data model. Our evaluations encompass various performance metrics, alongside investigations into multitask, hierarchical, and imbalanced learning for neural networks. Results: We introduce a novel metric, RE@R, tailored to the ICD coding task, which offers interpretable insights for healthcare informatics practitioners, aiding them in assessing the quality of assisted coding systems. Our findings highlight that selectively cherry-picking ICD codes diminish retrieval performance without performance improvement over the selected subset. We show that optimizing for metrics such as NDCG and AUPRC outperforms traditional F1-based metrics in ranking performance. We observe that Neural Network training on different ICD levels simultaneously offers minor benefits for ranking and significant runtime gains. However, our models do not derive benefits from hierarchical or class imbalance correction techniques for ICD code retrieval. Conclusion: This study offers valuable insights for researchers and healthcare practitioners interested in developing and evaluating CAC systems. Using a straightforward sequential neural network model, we confirm that medical prescriptions are a rich data source for CAC systems, providing competitive retrieval capabilities for a fraction of the computational load compared to text-based models. Our study underscores the importance of metric selection and challenges existing practices related to ICD code sub-setting for model training and evaluation.

show abstract

“…However, the challenges of achieving efficient hierarchical product classification automation lie in the vast volume of e-commerce data [5], the ambiguity in product descriptions, data imbalance [6], multilingualism [7], [8], and scalability [9]. Recent advancements in machine learning, along with the continuous efforts of numerous authors, provide tools to address these challenges.…”

Section: Introductionmentioning

confidence: 99%

A Flat-Hierarchical Approach Based on Machine Learning Model for e-Commerce Product Classification

Cotacallapa,

Saboya,

Canas Rodrigues

et al. 2024

IEEE Access

View full text Add to dashboard Cite

Within the e-commerce sphere, optimizing the product classification process assumes pivotal importance, owing to its direct influence on operational efficiency and profitability. In this context, employing machine learning algorithms stands out as a premier solution for effectively automating this process. The design of these models commonly adopts either a flat or local (hierarchical) approach. However, each of them exhibits significant limitations. The local approach introduces taxonomic inconsistencies in predictions, whereas the flat approach becomes inefficient when dealing with extensive datasets featuring high granularity. Therefore, our research introduces a solution for hierarchical product classification based on a Machine Learning model that integrates both flat and local (hierarchical) classification approaches using a 4-level electronic product dataset obtained from a renowned e-commerce platform in Latin America. In pursuit of this goal, a comparative analysis of seven machine learning algorithms, including Multinomial Naive Bayes, Linear Support Vector Classifier, Multinomial Logistic Regression, Random Forest, XGBoost, FastText, and Voting Ensemble, was conducted. This hybrid approach model exhibits superior performance compared to models using a single approach. It surpassed the top-performing flat approach model by 0.15% and outperformed the leading local approach (Local Classifier per Level) model by 4.88%, as measured by the weighted F1score. Additionally, this paper contributes to the academic community by presenting a significant Spanishlanguage dataset comprising over one million products and discussing the optimal preprocessing techniques tailored for the dataset. It also addresses the study's inherent limitations and potential avenues for future exploration in this field. INDEX TERMS machine learning, e-commerce, hierarchical product classification, local classifier per level, ensemble

show abstract

Handling imbalance in hierarchical classification problems using local classifiers approaches

Cited by 6 publications

References 68 publications

Learning Ordinal–Hierarchical Constraints for Deep Learning Classifiers

Learning Ordinal–Hierarchical Constraints for Deep Learning Classifiers

Creating a computer assisted ICD coding system: performance metric choice and use of the ICD hierarchy

A Flat-Hierarchical Approach Based on Machine Learning Model for e-Commerce Product Classification

Contact Info

Product

Resources

About