Akademik Veri Yönetim Sistemi
15th INTERNATIONAL CONFERENCE ON ENGINEERING & NATURAL SCIENCES, Muş, Türkiye, 4 - 06 Mart 2023, ss.95-105
Chronic Kidney Disease (CKD) is a life-threatening disease. Early diagnosis of mortal illnesses is of utmost
importance. Utilizing historical data for the prediction of future health conditions is the easiest, less costly and
non-invasive way. This approach yields to early detection of diseases as well. Machine Learning (ML) and
Deep Learning (DL) models are the most common technics for intelligent estimations. In this paper, we
proposed a new 1D-CNN deep learning model to predict CKD problems. The proposed model is compared
with K- Nearest Neighbour (KNN), Decision Tree (DT), Support Vector Ma- chine (SVM) and Random Forest
(RF) algorithms. Hyperparameters affect the model performances of ML and DL algorithms. In this study,
hyperparameter optimization is employed for all models using the Grid Search CV optimization algorithm. In
order to improve model quality, 10-fold cross-validation is conducted for all ML models. The comparison
metrics, accuracy, precision, recall and F1 score are utilized for performance evaluations. The accuracy metric
is compared for both training and test data sets. The remaining metrics, precision, recall and F1 score, are
calculated for only the test data set. The proposed 1D-CNN model achieved competitive results among all
machine learning models. We have reported 98.93% accuracy for both training and test data sets. Moreover,
98.93% precision is calculated for the test data set. The recall metric value is 98.93% for the test data set.
Finally, the F1 score is calculated as 98.93%, which is the best achieve- ment among all ML models in this
study.