Neural network optimization-based intelligent evaluation algorithm for international Chinese teaching effect

Evaluating the effectiveness of teaching Chinese as an international language is considered to be important within the context of global education; as such there is great difficulty in assessing whether or not the way it is delivered is effective due to variances in both learners and their respective learning environments. An intelligent evaluation algorithm based on neural network optimization to address teaching effectiveness problems is developed. The Chaotic Dragonfly Algorithm Optimized Multilayer Feed-Forward Neural Network (CDA-MFFNN) is implemented, to determine a higher level of prediction accuracy for effectiveness in teaching strategies. The data set used to train the CDA-MFFNN model is the Chinese Teaching Effectiveness Data Set, which contains 6,531 data records related to performance, engagement, and instructional attributes. Preprocessing involves min-max normalization and semantic coding of categorical variables, to reduce the number of dimensions and capture latent property patterns. While autoencoders are used as feature extraction method to capture latent patterns and reduce dimensionality. The MFFNN models the relationships between the processed features in a complex non-linear manner while the CDA optimizes the weights of the neural networks to improve convergence and prevent being stuck in local minima during training. Python-based experimental evaluation demonstrated a clear performance advantage over the baseline approaches, with the proposed method attaining 98.65% accuracy and a 98.20% F1-score. It also produced lower prediction errors, recording an MSE of 15.72 and an RMSE of 3.96. This research contributes to the development of intelligent educational analytics and provides action-oriented guidance for curriculum developers and policy makers. Not applicable.