Figure 4 With feature selection DT on Spambase dataset.
Related Figures (11)
Fig. 1: Proposed framework: Architecture and workflow for machine learning-based email ham and spam detection. 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. 5: Without feature selection DT on Spambase dataset 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. Fig. 6: Comparing classification metrics across varied algorithms.
