@article{article_1558835, title={Using Artificial Neural Network in the Prediction of Metacognitive Thinking Skills: Performance Evaluation of Network Structure and Training Methods}, journal={Karaelmas Fen ve Mühendislik Dergisi}, volume={15}, pages={107–118}, year={2025}, DOI={10.7212/karaelmasfen.1558835}, author={Barin Özkan, Sibel and Özkan, Yasin}, keywords={Artificial neural networks, machine learning, metacognition}, abstract={Metacognitive thinking skill is one of the most important skills that individuals should have today. In recent years, technology has been frequently utilized in the acquisition of this skill. The development and transformation of technology, especially robots in recent years, has affected societies both directly and indirectly. The predictions of scientists have started to find a response in the education community for many years. Therefore, it has become unthinkable that robots do not affect the field of education. In this study, a feed-forward back-propagation artificial neural network was used to predict metacognition training and planning, monitoring and evaluation skills. In this context, information on 250 instructors working in universities affiliated to the Council of Higher Education was used as the input of the artificial neural network. A suitable network design was obtained by making metacognition training prediction for different artificial neural network designs. Then, the network was trained using different training algorithms for this appropriate network design and the most appropriate training method was determined among these algorithms. The performances of different network designs and training algorithms were analyzed in terms of correlation coefficient and mean squared error. In terms of evaluating the parameters affecting the performance, the designs with 1 and 2 hidden layers were run 10 times, and the network structure with 20 and 10 hidden neurons in each layer, respectively, was determined as the most appropriate design. For this network structure, the best results in terms of performance parameters were observed when Conjugate Gradient Fletcher-Reeves and Levenberg-Marquardt training algorithm were used.}, number={1}, publisher={Zonguldak Bulent Ecevit University}