Evaluating Feature Selection Methods for Accurate Diagnosis of Diabetic Kidney Disease
Author Information
Author(s): Maeda-Gutiérrez Valeria, Galván-Tejada Carlos E., Galván-Tejada Jorge I., Cruz Miguel, Celaya-Padilla José M., Gamboa-Rosales Hamurabi, García-Hernández Alejandra, Luna-García Huizilopoztli, Villalba-Condori Klinge Orlando
Primary Institution: Unidad Académica de Ingeniería Eléctrica, Universidad Autónoma de Zacatecas
Hypothesis
The aim of this research is to compare distinct methodologies of feature selection to identify discriminant risk factors that may be beneficial for early treatment and prevention of diabetic kidney disease.
Conclusion
The proposed methodology has the potential to facilitate the prompt identification of nephropathy and non-nephropathy patients, and could be used in clinical settings as a preliminary computer-aided diagnosis tool.
Supporting Evidence
- Galgo with Random Forest achieved the best performance with only three predictors: creatinine, urea, and lipids treatment.
- The model displayed a moderate classification performance with an area under the curve of 0.80.
- The study emphasizes the importance of early detection and intervention in diabetic kidney disease.
Takeaway
This study looked at different ways to pick important health factors to help doctors find kidney problems in diabetes patients early.
Methodology
The study evaluated a dataset from 22 patients using four feature selection algorithms: Univariate, Boruta, Galgo, and Elastic net, followed by classification with a random forest model.
Potential Biases
The dataset may not represent the broader population due to cultural, genetic, and environmental factors unique to Mexico.
Limitations
The study was limited by the small sample size of 22 patients and may only be applicable to the Mexican population.
Participant Demographics
The participants included 22 Mexican patients diagnosed with type 2 diabetes, aged between 36 and 84 years.
Statistical Information
Confidence Interval
0.80 (±0.3535 SD)
Statistical Significance
p<0.05
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website