Academia.eduAcademia.edu

Figure 1 – uploaded by Joyece Jane

See full PDFdownloadDownload figure
Fig. 1: Proposed framework: Architecture and workflow for machine learning-based email ham and spam detection.

Figure 1 Proposed framework: Architecture and workflow for machine learning-based email ham and spam detection.

Related Figures (11)

Fig. 2: With feature selection ANN on Spambase dataset. As depicted in Figure 2, two filter feature selection methods, namely Pearson correlation and chi-squared, were applied for the neural network classification. Surprisingly, the perfor- mance metric score was found to be poor, with the accuracy dropping by 72% compared to the accuracy achieved without feature selection. This outcome leads us to conclude that for neural network algorithms, the feature selection technique utilized did not improve performance, but instead resulted in a significant decrease in accuracy.
Fig. 2: With feature selection ANN on Spambase dataset. As depicted in Figure 2, two filter feature selection methods, namely Pearson correlation and chi-squared, were applied for the neural network classification. Surprisingly, the perfor- mance metric score was found to be poor, with the accuracy dropping by 72% compared to the accuracy achieved without feature selection. This outcome leads us to conclude that for neural network algorithms, the feature selection technique utilized did not improve performance, but instead resulted in a significant decrease in accuracy.
Fig. 3: Without feature selection ANN on Spambase dataset.
Fig. 3: Without feature selection ANN on Spambase dataset.
Fig. 4: With feature selection DT on Spambase dataset.
Fig. 4: With feature selection DT on Spambase dataset.
Fig. 5: Without feature selection DT on Spambase dataset
Fig. 5: Without feature selection DT on Spambase dataset
TABLE I: Experimental results
TABLE I: Experimental results
TABLE III: Comparative analysis with other studies Particularly striking is the discernible impact on the different datasets, complemented by preprocessing and feature selec- tion, achieving remarkable accuracy scores that owe their success to the strategic implementation of feature selection and data preprocessing techniques in the proposed framework.
TABLE III: Comparative analysis with other studies Particularly striking is the discernible impact on the different datasets, complemented by preprocessing and feature selec- tion, achieving remarkable accuracy scores that owe their success to the strategic implementation of feature selection and data preprocessing techniques in the proposed framework.
Fig. 6: Comparing classification metrics across varied algorithms.
Fig. 6: Comparing classification metrics across varied algorithms.
Related topics:
Machine Learning
580 California St., Suite 400
San Francisco, CA, 94104
© 2025 Academia. All rights reserved