The Impact of Balancing Techniques and Feature Selection on Machine Learning Models for Diabetes Detection

Fırat Üniversitesi Mühendislik Bilimleri Dergisi, Journal Year: 2025, Volume and Issue: 37(1), P. 303 - 320

Published: Jan. 24, 2025

The detection of diabetes is crucial for effective management and prevention the disease, which poses significant health risks globally. This study introduces a novel approach to by combining advanced data balancing techniques feature selection methods, including Lasso (L1) regularization, enhance performance predictive models in imbalanced datasets. Techniques such as Random Under Sampling (RUS), Adaptive Synthetic (ADASYN), Minority Over-sampling Technique (SMOTE) were employed alongside Forest (RF), CatBoost (CB), Extreme Gradient Boosting (XGB), K-Nearest Neighbors (KNN), Gaussian Naive Bayes (GNB), Logistic Regression (LR), (GB) assess their impact on model accuracy generalization capabilities. findings reveal that RF achieved highest 93.25% when utilizing SMOTE technique, underscoring importance appropriate handling strategies improving outcomes. Furthermore, all features utilized without selection, attained an 95.31%, indicating model’s capacity capture complex patterns richness maximized. comprehensive methodology used higher than research literature provided important outputs developing reliable prediction healthcare.

Language: Английский

Application of Transformer-Based Deep Learning Models for Predicting the Suitability of Water for Agricultural Purposes

Water, Journal Year: 2025, Volume and Issue: 17(9), P. 1347 - 1347

Published: April 30, 2025

Water is the most vital component for sustainability of living beings on Earth. From plants to human beings, every single being Earth needs water its survival. In this research, a novel model has been developed in order predict suitability agricultural purposes. This research ALBERT Base v2 detecting quality and named Potability Detection (ALBERT-WPD) model, customized from transformer model. The was tested using dataset Kaggle, performance evaluated. used ten parameters. both models measured metrics, accuracy, precision, recall, F1-score. traditional (CNN RNN) were compared against measure efficiency potability prediction. findings revealed that gained higher accuracies than models: 91% altered ALBERT-WPD rendered 96% accuracy. classification results (precision, F1-score) obtained class 93%, 98%, those non-potability 95%, 96%, respectively. study found detection procures accuracy with optimization method. concludes (BERT-based) (>95%) fewer parameters comparison which utilize more show exhibit rapid data processing handle large datasets efficiently; handling such complicated when models, as they have vanishing gradient encounter temporal loss challenges. Thus, significance proposed dwells within use “transformers” an advanced machine learning quality, showing transformers are future learning.

Language: Английский