To better improve student performance and make students more receptive to knowledge, a study of artificial intelligence algorithms for the development and evaluation of college teachers’ teaching skills is proposed. Connections are established with multiple neurons in the periphery through dendrites and axons, and weighted integration of neuronal inputs to the next neuron, given an intelligent learning target. The sensitivity is back-propagated in the adjustment of model parameters, and the centroids are selected to calculate the Euclidean distance to obtain an accurate prediction model, and the weight parameters and bias vectors of the model are adjusted. Using a two-by-two comparison to determine the hierarchical factors, the weights of the relative importance of all factors at each level were calculated and ranked, and the multi-level index system was formed according to the constraint relationship between factors. According to the hierarchical structure of the teaching ability of college teacher educators and the opinions of some experts, the weights of each element in each matrix were determined to reasonably evaluate the effective teaching ability of college teachers in the classroom. The analysis results show that the artificial neural network model has a relative error value of about 1.5% and high numerical accuracy in evaluating teachers’ classroom teaching ability by hierarchical analysis.