摘要
为了有效处理企业越来越多的业务数据,为企业业务提升和用户价值挖掘提供积极帮助,将机器学习方法应用于某运营商客户业务数据处理过程。首先对原始数据进行预处理,去除重复值、缺失值、异常值,并进行标准化处理,然后对不平衡数据采用合成少数类过采样(synthetic minority over-sampling technique,SMOTE)技术进行过采样,减少了预测的偏差。对处理后数据分别建立传统神经网络模型、优化神经网络模型和随机森林模型,并通过结构调优和参数调优等进行模型优化,对运营商客户进行预测与分析。结果表明,优化后的模型预测准确率可达96%,有良好的客户预测与分析效果,可见优化模型的有效性。最后为运营商挽留流失客户、维系非流失客户提供了解决方案,为运营商实施精准营销、节省运营商营销成本和创造更多利润提供了技术支持。
In order to effectively processing more and more business data of enterprises and provide active help for business promotion and user value mining,machine learning methods were applied to the business data of a certain operator.First,the original data was preprocessed to remove duplicate values,missing values,outliers,and be standardized.Synthetic minority over-sampling technique(SMOTE)was used to oversample the unbalanced data to reduce the forecasting deviation.Then established traditional neural networks model,optimized neural networks model,and random forest model for the processed data,and optimized the models through structure tuning and parameter tuning to predict and analyze the operator’s customers’churn.The results show that,the optimized model’s prediction accuracy rate can reach 96%.It has good effects of customer forecasting and analysis.The results proves that it is an effective model.Solutions are provided to retain lost customers and maintain non-churning customers for the operator,and technical support are provided for operators to implement precision marketing,save operators'marketing costs and create more profits.
作者
周艳聪
郝园媛
ZHOU Yan-cong;HAO Yuan-yuan(School of Information Engineering, Tianjin University of Commerce, Tianjin 300134, China)
出处
《科学技术与工程》
北大核心
2022年第2期585-592,共8页
Science Technology and Engineering
基金
国家自然科学基金(71572125)。
关键词
客户行为分析
神经网络
随机森林
机器学习
customer behavior analysis
neural network
random forest
machine learning
作者简介
第一作者:周艳聪(1978-),女,汉族,河北饶阳人,博士,教授,研究方向:智能信息处理、机器学习,E-mail:zycong78@126.com。
