用周期模型和近邻算法预测话务量时间序列

客服中心话务量虽然具有周期性,但在不同时间遵循不同变化规律,这是话务量预测的难点。针对这个问题,以某电信公司一年的实际话务数据为基础,分别采用周期模型和基于实例的近邻算法进行话务量时间序列预测,并对比分析了两种预测方法的效果。实验数据表明,对工作日话务量的预测,周期模型的预测效果优于近邻算法;对非工作日话务量的预测,近邻算法的预测效果优于周期模型。为取得更好的预测效果,实现了周期模型和近邻算法相结合的预测方法。结果表明,在最好的情况下,该方法的预测精度比周期模型提高约19.7%,比近邻算法提高约48.8%。

Traffic Time Series Prediction Based on Periods-Model and Case-Based Nearest Neighbor Algorithm

Although the traffic has a characteristic of periodicity, it conforms to different rules at different time, and this is the difficulty of prediction. Aiming at this problem and based on the true data of a certain company, the time series prediction by periods-model and case-based nearest neighbor algorithm was proposed, then their results are compared. The experiment demonstrates that periods-model performs better in predicting weekdays, and case-based nearest neighbor algorithm performs better in predicting weekends. For better prediction results, combination of periods-model and case-based nearest neighbor algorithm ware realized. The experiment demonstrates that at the best, the prediction precision is improved by about 19.7% using this method than periodsmodel, and is improved by about 48.8% than nearest neighbor algorithm.