With the advent of the Internet plus era, digitalization and informatization have penetrated into various fields, and have had a profound impact. At the same time, it has also brought great changes to the concept of education. In data mining, there are many clustering technologies, among which K-means clustering (K-Means) algorithm is widely used because of its simple process and good convergence performance. This paper mainly made some improvements on the traditional K-Means algorithm, which not only maintains the advantages of the K-Means algorithm, but also overcomes the shortcomings of the initial cluster center (ICC) dependence. Applying the improved K-means algorithm to digital teaching can improve the arrangement of digital teaching resources (DTR) and further promote the application of digital teaching. In this paper, 300 college students and 20 college teachers were investigated by questionnaire, and 300 students were divided into groups for a half year comparative experiment. Through the comparative analysis of the teaching effects before and after the experiment, it can be seen that before the experiment, the interest of the experimental group in learning mathematics and the mathematical scores were 4.8 and 4.1 respectively. After the digital teaching, the scores of the two groups were 7.6 and 7.9 respectively, while the scores of the control group before and after the traditional teaching were not significantly different. To sum up, the digital teaching of K-Means algorithm would be improved, which has a positive impact on the teaching effect.