Research On MTS Based On Improved Measurement Tool And Threshold Calculation Method And Its Application In Email Filtering

With the development of the internet and mobile terminals, email being a very important communication tool gradually. While numerous spams bring many challenges for clients and service providers, research about email filter caused widespread attention recently. Mahalanobis Taguchi system (MTS) aim to recognize models and classify for multi-dimensions problems, this method don’t need any hypothesis about input data and could effective reduce key variables before classification. This work tries to improve traditional MTS in measurement tool and threshold calculation method by different methods, apply the improved MTS to email filtering. This dissertation includes three points:(1) Research on new measurement tool by grey incidence model. Metrics means the similarity between single samples and collective, it also decides the sample’s sort. MTS view Mahalanbis distance as the measurement tool has consider attributes correlation compare with Euclidean distance, however, Mahalanbis distance just weigh the distance from the sample to the whole and ignore the curve similarity in spatial extent. Grey incidence model is a new tool to calculate the curve similarity for samples, which has good generality. To compute the similarity comprehensive and improve accuracy, this work constructs a new measurement tool by combination the Mahalanobis distance and grey incidence model in the way of leaner weight.(2) Research on threshold by Receiver Operating Characteristic (ROC) curve. Threshold compute methods in MTS has been discussion constantly, exist methods have different degree faults and few applications. ROC curve used to analysis classify effect and calculate system threshold major in medical diagnosis field. ROC curve been used to compute the threshold for MTS in this paper, which could let the process much visible and increase the accuracy.(3) Study on email classification based on improved MTS. As the result shows:the new MTS is much better than traditional MTS in every index, those improve methods is valuable; compares with other widespread used email predict methods, the improved MTS achieve higher accuracy, lower cost, and more effective by save key attributes.