Research And Implementation Of The Model For Detecting Abnormal Correlation Data

In recent years,computers have further improved their ability to generate,collect,process,and analyze data.The concept of big data has also gradually penetrated the hearts of the people,and all walks of life have recognized the importance of data and the value contained in the data.Data mining has also been gradually studied by scholars.The abnormal warning in data mining has great significance for reality.The traditional Cumulative Sum(CUSUM)anomaly discovery model was used in a provincial border system,but false positive of the early warnings reported by the model during the application process were very high,and the analysis showed that the model made early warning through the fluctuation of the data.The greater the fluctuation,the greater the probability of early warning,but the fluctuation of the data was normal in the actual situation,so the model could make an early warning,but the early warning results obtained were inconsistent with the actual situation,and the false positive rate was relatively large.Based on the data provided by the border epidemic alert detection system,the data were pre-processed to obtain the daily data of cough and sore throat as the experimental data of this paper.Then built a correlation model and substituted the data of cough and sore throat to get a strong correlation between cough and sore throat.Next the text proposed a smooth correlation coefficient,and the correlation coefficient sequence of cough and sore throat changes with time was calculated,and the sequence was used as the experimental data of the abnormal discovery model.In this paper,the CUSUM model was built,and the above experimental data was processed separately in the form of inserting and replacing,and the data with abnormality was obtained.This data was then substituted into the CUSUM model for several experiments.In the end,CUSUM had a higher detection rate for the replacement data than the inserted data,but the detection rate was not very high and needed to be improved.Then the CUSUM defect was obtained through analysis.Based on the defect,a data anomaly discovery model(DEW and DWT Extremely Find Model,DEFM)based on wavelet transform and dynamic time warping was proposed.After the DEFM model was built,the detection and DEF detection rates were also verified by inserting and replacing,and compared with the detection rate of CUSUM.When the length of the inserted and replaced data was small,it generally belonged to the normal fluctuation in the data(also known as sporadic anomaly).At this time,the abnormal detection rate of the CUSUM model was higher than the DEFM model,indicating that the false positive rate of CUSUM was higher.When the length of the inserted and replaced data was long,it was generally an abnormal fluctuation of the data(non-incidental anomaly),and the detection rate of the DEFM model was higher than that of the CUSUM model.