Study On The Improvement Of The CMCC Business Monitoring Method Based On Machine Learning

Along the occurrence of mobile Internet, the business such as cellphone game, cellphone video, cellphone supermarket, wireless music, cellphone reading and cellphone animation occurs also. At the same time, the data business of China Mobile develops rapidly, whichmakes the difficulty and complexity of daily business management become larger.Due to loss to the company caused by wrong orders, vicious consumption and other problems, the customer’s satisfaction degree reduces.Therefore, it requires some powerful means to monitor the data business so as to find the problem in time.Currently, the business monitoring system in operating mainly aims at monitoring and analyzing the phenomenon of sharply-increased data size caused by customer’s vicious consumption, large wrong order quantity, illegal profit earning by system bug,but it has the problem of reporting the situations by mistake or leakage.For example, cellphone game, cellphone supermarket, cellphone reading and other monthly-payment business are ordered at the beginning of month, which will cause the ordering quantity to have large fluctuation.Such problem belongs to the normal phenomena, but as the current system has the disadvantage in algorithm, it generates many wrong alarms. Aiming at the reporting by mistake or leakage in current business monitoring system, the paper carries out study, introduces new algorithm to solve the problem of reporting by mistake or leakage, trains the intelligent alarm filter by machine learning algorithm, and guides the alarm filtration through the alarm replying messages so as to relieve the staff’s workload. The paper has the following tasks:(1)Aiming business with large fluctuation in business quantity, it introduces DBSCAN clustering algorithm to solve the influence of little outburst data on the algorithm result and reduce the wrong alarm to data belong to the normal phenomena. Meanwhile, the paper improves the algorithm to improve its efficiency so that such algorithm can be applied to large-data business with precise time granularity.(2)Aiming periodic data traffic, the paper advances the monitoring method by applying neural network algorithm and Holt-Winters combined model to solve the problem of reporting leakage in abnormal data of periodic business. It improves the accuracy and recall ratio of alarm reported by the business monitoring system.(3)The alarm messages generated by the business monitoring system also needs the staff to handle. In order to reduce workload and improve the accuracy of reporting alarm by the business monitoring system, the paper advances to utilize the alarm replying message, train the alarm filter by data classification technique and guides the alarm filtration. The experiment shows it can separate the invalid alarm messages by the alarm filter, which reduces the workload.(4)Through improving the original algorithm of the business monitoring system, the paper can solve the problems of reporting alarm by mistake or leakage better, separate the invalid alarms by the alarm filter to realize the aim of reducing the workload. In the following, it will have deep study on the alarm filtering system to improve the alarm filtering effect through realizing its self-study and self-upgrading.