The quality assurance of education and training is of paramount importance, and the monitoring, assessment and evaluation process is indispensable for this purpose. A key determinant of the effectiveness of education is assessment, which serves not only to measure performance but also to support the learning process. The process of assessment and evaluation must take into account the diverse needs and abilities of students. Differentiated assessment supports individual learning paths and personal development. Assessment must be objective and valid, but it can achieve its deeper developmental purpose when personalized. Our research aims to enhance pedagogical assessment efficiency by developing a machine learning-supported assessment system. The research focuses on developing STEM skills and analyzes children's questions to understand their interests and needs better. During the research, we observe the characteristics of questions posed by children and develop a machine learning algorithm that helps provide suggestions to educators for creating assessment questions. Based on the data, we identify children's areas of interest and the difficulty level of questions. According to the research results, children's questions can significantly differ from question sets prepared by educators, suggesting that children's diverse perspectives and interests should be given even greater consideration in teaching. The research also raises further directions for development. It aims to use machine learning to understand better the diverse ways of thinking and differences in expression among children, thereby making education and assessment more effective. This can contribute significantly to the efficiency and effectiveness of the teaching-learning process.