Classifying Students' Grounded Mental Models on Energy with Deep Neural Network

Abstract

Abstract The concept of energy, a common interdisciplinary concept, is frequently used in daily life and can be associated with different subjects in terms of its scope. Additionally, it has an important place in science education, throughout primary, secondary and higher education. Thus, many grueling applications are carried out to detect learning situations. However, in recent years, opportunities have emerged to determine learning situations with deep learning networks (DNNs), which are a subunit of artificial intelligence. This study aimed to demonstrate the usability of DNNs in the classification of learning and to establish an example in this field of educational research concerning the concept of energy. To this end, a learning situation test was used to evaluate “energy types”, “transformation of energy” and “conservation of energy” to determine the “grounded mental model” (GMM). The test was used to determine the GMM of preservice teachers and to test the classification success with the least error by using a DNN. In this context, DNN models consisting of different parameters were designed for the training of deep neural networks. The models were analyzed with the most appropriate algorithm considering the number of hidden layers, the number of neurons in the hidden layers, the activation function, the optimization algorithm, the loss function, and the epoch values. Two methods were used for training and testing the ensemble classifiers and individual classifiers. The first is to divide the dataset into 70% training data and 30% test data, and the second is a 5-fold cross-validation method. The results were evaluated using the recall (R), specificity (S), accuracy (ACC), and F1 score metrics. According to the test results obtained from both methods, the ensemble classifier has the best classification performance. According to the results of the analysis with deep neural network algorithms, 95% classification accuracy was obtained. Trainers can use the designed DNN models as a validation tool in the detection of GMM.