Akademik Veri Yönetim Sistemi
Expert Systems with Applications, cilt.281, 2025 (SCI-Expanded, Scopus)
Classification tasks categorize data into predefined classes. Conventional methods, such as decision trees and support vector machines, are used for classification and have been widely adopted due to their effectiveness in handling various types of structured data, but in recent decades Convolutional Neural Networks (CNNs) rival them in computer vision. This study presents a novel CNN-based algorithm expanding classification tasks beyond images. The proposed method enables the transformation of each row within a pre-existing feature matrix into a representative image. Subsequently, a custom-designed CNN architecture is employed to classify these constructed images. The aim is to generalize CNN for any classification problem, providing competitive performance. Twelve datasets from the UCI repository are utilized to derive performance metrics from nine well-known classifiers. Hyperparameter tuning is employed on the constructed models to attain optimal performance results for each dataset. Ten times ten-fold cross-validation ensures reliable model estimates. The proposed method yields competitive performance values, even without hyperparameter tuning and by using a very simple CNN architecture. With the innovative methodology introduced in this study, achieving superior classification performance is feasible through hyperparameter tuning, feature extraction, feature selection, and transfer learning.