Design of an artificial intelligence supported student recognition system and the effect of differentiated instruction on conceptual learning

This thesis aims to investigate the effect of differentiated teaching practices integrated with artificial intelligence in order to effectively structure conceptual learning in science education. In this regard, the enrichment of differentiated teaching practices through educational data mining, learning analytics, and machine learning algorithms is targeted. Artificial intelligence tools were used to analyze learning styles, conceptual learning, and misconceptions. This study was conducted using a design-based research method. Three designs were implemented in the design-based research process. First, "Christou's Five-Step Model" was used to determine the learning styles of teacher candidates. Second, the "Learning Analytics Cycle Model" was used in the process of identifying the misconceptions of teacher candidates. In the third stage, the "ADDIE Design Cycle Model" was used to develop activities in differentiated instruction. The thesis study was conducted with three different sample groups in a design-based research process. The first sample group consisted of 522 volunteer teacher candidates studying in science education programs at eight different universities who were included in the process of determining AI-supported learning styles. The second sample group consisted of 510 volunteer teacher candidates studying in science education programs at 8 different universities who were included in the process of identifying misconceptions related to electricity supported by artificial intelligence. The third sample group consists of 30 teacher candidates enrolled in the science education program at Kırıkkale University, who were selected through purposive sampling and included in the differentiated teaching process. The thesis study primarily aimed to determine the learning styles of teacher candidates using artificial intelligence algorithms instead of human experts. In the studies conducted during this process, qualitative data was collected using the 'KDD Data Mining Process Model'. This data was analyzed using the Thematic Analysis Method. Natural Language Processing was used to process the dataset obtained from a total of 552 teacher candidates. The data was classified using artificial intelligence tools, namely Machine Learning and supervised learning algorithms. It was determined that the Machine Learning algorithm with the highest accuracy rate used to predict the learning styles of teacher candidates was the 'Ensemble Model' developed by the researchers, which was created from a combination of Multi-Layer Perceptron, Random Forest, and Decision Tree algorithms. At this stage, it was concluded that Natural Language Processing applications are important in the process of automatically determining learning styles and that thematic analysis is an innovative method in artificial intelligence applications. Furthermore, based on the quantitative data obtained within the scope of the study, the kappa (κ) value was analyzed to determine the meaningful difference between the artificial intelligence algorithm and the human expert assessment in the process of determining the learning styles of teacher candidates, and it was found that the level of agreement was 'almost perfect'. Secondly, in the thesis study, the artificial intelligence data of 510 candidates obtained on the subject of electricity was created using the Learning Analytics Cycle and analyzed using the thematic analysis method. In addition, the questions included in the web form developed by the researchers were used to determine the misconceptions and conceptual learning of teacher candidates. Frequency distribution was used to determine misconceptions. The data revealed that teacher candidates had 249 sentences containing misconceptions about electricity. The identified misconceptions were found to be related to those mentioned in the literature. The thematic analysis determined that the artificial intelligence algorithm with the highest performance rate was the triadic Ensemble Model developed by the researchers for each question. Furthermore, since the pre-test and post-test conceptual learning level data obtained from 30 teacher candidates during the experimental process of the thesis study were found to be normally distributed, they were analyzed using the dependent t-test. Finally, it was found that there was a statistically significant difference in favor of the final test on the conceptual learning level scores of teacher candidates under the influence of differentiated instruction. This thesis study concluded that differentiated instruction is effective in the conceptual learning levels of teacher candidates. Furthermore, it is noteworthy that the success rate of identifying misconceptions using artificial intelligence was quite high. It is believed that the findings will contribute significantly to the relevant literature and to those who will conduct similar research.