Improving imbalanced predictions via a novel watershed-based oversampling method: Application to sickle cell disease - Archive ouverte HAL Accéder directement au contenu
Pré-Publication, Document De Travail (Preprint/Prepublication) Année : 2023

Improving imbalanced predictions via a novel watershed-based oversampling method: Application to sickle cell disease

Yamna Ouchtar
  • Fonction : Auteur
  • PersonId : 1258370
Christian Kassasseya
  • Fonction : Auteur
  • PersonId : 1258371
Pablo Bartolucci
  • Fonction : Auteur
  • PersonId : 1258372

Résumé

Most of the time, real-world data sets are composed of classes with the same number of samples. But sometimes, for example in the case of fraud detection or rare disease diagnosis, the resulting dataset is composed of asymmetric classes. These datasets are called imbalanced datasets. Classification problems based on imbalanced datasets lead to errors and high variability. Thus, methods to deal with have been developed. In this paper, we propose a novel oversampling method, called WSSMOTE, based on the watershed transformation. We demonstrate that WSSMOTE improves prediction scores in some real-world datasets. In addition, our main goal is to improve prediction scores of an imbalanced dataset composed of sickle cell disease (SCD) biomarkers. SCD is a seriousinherited disease, and patients with SCD can be affected by vaso-occlusive crises, which are the main cause of hospitalization. During these hospitalizations, acute chest syndrome (ACS) is the leading cause of death. ACS occurs in approximately 20% of hospitalized patients. During two prospective studies (PRESEV1, 247 patients and PRESEV2, 393 patients) a predictive score for ACS based on clinical and biological data has been developed. The obtained negative predictive value (NPV) is high and relatively similar in both studies (98.9% and 94%, respectively) but, the obtained positive predictive value (PPV) is low and highly variable (44.7% and 27.9%, respectively). Hence, we want to improve prediction performance while reducing its variability. We demonstrate that existing oversampling methods fail to improve the PPV value, whereas WSSMOTE succeed in doing so. With WSSMOTE, the PPV increases from 24.6% to 28.9%, while maintaining a high NPV (96.6%). In addition, the overfitting of the PPV value is reduced from 13.3 % to 1.2.
Fichier principal
Vignette du fichier
WSSMOTE_Elsivier_V2.pdf (727.29 Ko) Télécharger le fichier
Origine : Fichiers produits par l'(les) auteur(s)

Dates et versions

hal-04112987 , version 1 (01-06-2023)

Identifiants

  • HAL Id : hal-04112987 , version 1

Citer

Yamna Ouchtar, Benjamin Perret, Christian Kassasseya, Pablo Bartolucci, Laurent Najman. Improving imbalanced predictions via a novel watershed-based oversampling method: Application to sickle cell disease. 2023. ⟨hal-04112987⟩
87 Consultations
57 Téléchargements

Partager

Gmail Facebook X LinkedIn More