Academia.eduAcademia.edu

Figure 1 – uploaded by Srikanth Kadainti

See full PDFdownloadDownload figure
Fig. 1(a). Distribution of Z2 Scores [2a]

Figure 1 (a). Distribution of Z2 Scores [2a]

Related Figures (7)

List 1. Classification errors Spam Detection by Combining Bayesian Method and Regression Analysis In this paper, we re-examine the Naive Bayesian filter by identifying some important statistical features of the filter. We also report the results of our experiment with Enron data set and carry out statistical analysis and establish the merits of the new method.
List 1. Classification errors Spam Detection by Combining Bayesian Method and Regression Analysis In this paper, we re-examine the Naive Bayesian filter by identifying some important statistical features of the filter. We also report the results of our experiment with Enron data set and carry out statistical analysis and establish the merits of the new method.
Table 1. Partial list of tokens and their probabilities obtained from the training data [2a]
Table 1. Partial list of tokens and their probabilities obtained from the training data [2a]
Table 3. Z Scores for different mails [2a]
Table 3. Z Scores for different mails [2a]
Table 4. Summary measures of Z2 statistic for naive bayesian and modified naive bayesian methods [2a] Interestingly the misclassification percent has come down to 8.76% and the FP has drastically come down. False negatives however, have increased [2a]. The new model has 67% of sensitivity and 98% of specificity. The odds ratio is also very high in this case compared to the results given in Table 4. The AUC indicates that the probability is 0.977 that a randomly selected spam mail has Z2 value higher than that of a randomly selected ham mail [2a]. The results of this modified Bayesian procedure classification as shown in Table 4.
Table 4. Summary measures of Z2 statistic for naive bayesian and modified naive bayesian methods [2a] Interestingly the misclassification percent has come down to 8.76% and the FP has drastically come down. False negatives however, have increased [2a]. The new model has 67% of sensitivity and 98% of specificity. The odds ratio is also very high in this case compared to the results given in Table 4. The AUC indicates that the probability is 0.977 that a randomly selected spam mail has Z2 value higher than that of a randomly selected ham mail [2a]. The results of this modified Bayesian procedure classification as shown in Table 4.
Table 5. Comparison of performance statistics under random sampling [2a]
Table 5. Comparison of performance statistics under random sampling [2a]
Fig. 2. Percentage misclassification during random testing [2a] Research Updates in Mathematics and Computer Science Vol. 3 pam Detection by Combining Bayesian Method and Regression Analysis Spam Detection by Combining Bayesian Method and Regression Analysis
Fig. 2. Percentage misclassification during random testing [2a] Research Updates in Mathematics and Computer Science Vol. 3 pam Detection by Combining Bayesian Method and Regression Analysis Spam Detection by Combining Bayesian Method and Regression Analysis
Related topics:
Classification Models
580 California St., Suite 400
San Francisco, CA, 94104
© 2025 Academia. All rights reserved