Bengali textbooks used specific scientific terms that are difficult to understand for students and especially those in the low-asset learning environment. Thus the main objective of this paper will be to provide classification of Bengali science terms based on the frequency of books for every school grades. In this case, the term definitions dataset of 385 terms were collected from textbooks by term extraction and annotation. To this purpose, different modelling techniques such as machine learning classifiers, deep learning models, and transformer-based pretrained language models were employed. Based on the accuracy: Bangla-BERT-Base is being the most accurate model with an accuracy of 97%. While the traditional methods offered strong benchmarking solutions, the CNN came out as the strongest deep learning approach. ChatGPT and other generative AI models were also tested in the work in zero-shot and few-shot settings but they also did not show remarkable results. The conclusions make it possible to propose further tools for automatic classification of complexity and serve as a basis for creating educational aids for the further simplification of an extended list of terms for Bengali students.