In order to overcome the disadvantage of the canonical artificial bee colony algorithm, which has low search efficiency and slow convergence, an improved artificial bee colony algorithm is proposed. This algorithm increases the convergence speed by introducing the arithmetic crossover operation and guiding the search direction by the global best solution. The proposed algorithm is proved to be effective with a test on seven benchmark functions. On the basis of previous work, according to the drawbacks of the K-means algorithm, the K-means algorithm based on the improved artificial bee colony algorithm is proposed, and the function of automatically selecting the best number of clusters is added. A test on the artificial data sets and UCI real data sets verifies the performance of the proposed algorithm.