Research on personalised college English teaching technology based on deep feature extraction and scene feature mining

Determining potential applications for personalized college English instruction through deep feature extraction and scene feature mining is essential for enabling Deep Learning assisted AI systems to generate effective educational outcomes. Traditional instructional approaches often struggle within dynamic digital environments, as they overlook variations in students’ learning styles, engagement levels, and behavioral patterns. This research proposes a hybrid methodology that integrates Convolutional Neural Networks for deep audio feature extraction with Random Forest algorithms for contextual and scene feature mining. The objective is to construct a personalized learning environment capable of adapting dynamically in real time to individual learner behaviors and preferences. A hybrid deep learning model is introduced to predict teaching effectiveness using speech related features such as clarity, rhythm, and engagement level. Feature fusion is employed by combining audio based deep features with contextual scene features to classify teaching effectiveness into three categories: low, medium, and high. Experimental results demonstrate strong predictive performance, achieving an accuracy of 0.9900, precision of 0.9906, recall of 0.9907, and an F1 measure of 0.9905, with minimal performance fluctuations. This AI assisted instructional framework enables college English teachers to enhance teaching quality by receiving immediate, data driven feedback, ultimately fostering improved student centered learning experiences.