The performance of the proposed SVM-PSO was tested. According to previous studies, a confusion matrix is the most popular way of assessing the effectiveness of ML models. In addition to a standard confusion matrix, we also used a swarm search graph in our evaluation.Table 6 shows the accuracy achieved with each dataset in experiment 1. SVM achieved an accuracy of 0.82 with the ALL-IDB1&2 datasets and 0.79 with the online dataset24. In comparison, our hybrid SVM-PSO algorithm increased the accuracy to 0.97 with the ALL-IDB1&2 datasets and 0.92 with the online dataset. There was a clear difference in the accuracy and confusion matrix between the hybrid model (SVM-PSO) and the SVM algorithm alone, with the hybrid model achieving better accuracy and a higher detection rate.Table 6 Experiment 1 results.Furthermore, the accuracy of each model was tested with each dataset using a confusion matrix, which visualizes the classification performance of the test dataset, indicating the presence of positive and negative instances as classified by the model’.Figure 7 shows the confusion matrix results for SVM with the online dataset. The TF value for the online dataset of 49 means that there were 49 records with an actual and predicted negative value. Moreover, there were 50 records with an actual and predicted positive value. Therefore, the correct classification was made for 99 of 125 records. Furthermore, the TF value for the ALL-IDB1&2 datasets was 43 (Fig. 8), meaning that there were 43 records with an actual and predicted negative value. Since there were 60 records with an actual and predicted positive value, the correct classification was made for 103 records of 125 records.Fig. 7Confusion matrix of SVM with online dataset, experiment 1.Fig. 8Confusion matrix of SVM with ALL-IDB 1&2, experiment 1.Next a confusion matrix was applied to our suggested hybrid model (SVM-PSO). As shown in Fig. 9, for the online dataset the TF value was 60, meaning that there were 60 records with an actual and predicted negative value. Moreover, there were 55 records with an actual value and predicted positive value. Therefore, correct classification was made for 115 records of 125 records. For the ALL-IDB 1&2 datasets the TF value was 63, meaning that there are 63 records with an actual and predicted negative value (Fig. 10). Moreover, there were 59 records with an actual and predicted positive value. Thus, the correct classification was made for 122 of 125 records (Fig. 10).Fig. 9Confusion matrix of hybrid (SVM-PSO) model with the online dataset, experiment 1.Fig. 10Confusion matrix of hybrid (SVM-PSO) model with the online dataset, experiment 1.The comparison showed that the hybrid SVM-PSO model for ALL detection has an obvious benefit on disease detection accuracy, increasing the number of correctly classified datasets, as shown in the confusion matrices. There was a clear difference in accuracy between our hybrid model and the SVM algorithm alone, as the hybrid model achieved higher accuracy and detection rates. Overall, SVM-PSO exhibited better accuracy than the stand-alone algorithm. With the online dataset, the SVM algorithm achieved an accuracy of 79%, while SVM-PSO achieved a much higher accuracy of 92%. We found the same result with ALL-IDB 1&2, as our proposed hybrid model achieved a higher accuracy of 97%.We used swarm search graphs to further assess the performance of SVM-PSO in experiment 1.As shown in Fig. 11 the particle swarm search behavior in iteration 8 resulted in the optimal global solution, and as shown in Fig. 12, the best global solution was found in iteration 52. The results for experiment 2 are presented in Table 7.Fig. 11The Swarm search in iteration 8.Fig. 12The Swarm search in iteration 52.Table 7 Experiment 2 results.As shown in Table 7, SVM-PSO exhibited higher detection accuracy than SVM alone in experiment 2. Four experiments were conducted to investigate the performance of the hybrid model. SVM achieved an accuracy of 0.70 with the online dataset produced and 0.74 with the ALL-IDB 1&2 datasets. Meanwhile, SVM-PSO achieved an accuracy of 0.87 and 0.93 in the two datasets, respectively. Thus, the hybrid model achieved more accurate results. Confusion matrices were used to assess the accuracy of the models as in Figs. 13 and 14.As shown in Fig. 13, the TF value was 32 when using SVM with the online dataset, meaning that there were 48 records with an actual and predicted negative value. Since there were 57 records with an actual and predicted positive value, the correct classification was made for 105 of 150 records.The confusion matrix presented in Fig. 14 shows the results for the SVM algorithm with the ALL-IDB 1&2 datasets. The TF value for ALL-IDB 1&2 was 58, meaning that there were 58 records with an actual and predicted negative value. Moreover, there were 111 records with an actual and predicted positive value. Thus, the correct classification was made for 111 of 150 records.Fig. 13Confusion matrix of SVM with the online dataset, experiment 2.Fig. 14Confusion matrix of SVM with the ALL-IDB 1&2 dataset, experiment 2.The SVM-PSO results with the online dataset result are shown in the confusion matrix in Fig. 15. The TF value was 69, meaning that there were 69 records with an actual and predicted negative value. Moreover, there were 61 records with actual and predicted positive values. Hence, the correct classification was made for 130 of 150 records. Meanwhile, the SVM-PSO results with ALL-IDB 1&2 are shown in Fig. 16. The TF value was 73, indicating that there were 73 records with an actual and predicted negative value. Moreover, there were 67 records with an actual and predicted positive value, meaning that the correct classification was made for 140 of 150 records.Fig. 15Confusion matrix of hybrid the model with the online dataset, experiment 2.Fig. 16Confusion matrix of hybrid (SVM-PSO) model with the ALL-IDB 1& 2IDBdataset, experiment 2.The results from the second experiment showed that SVM-PSO still had the highest accuracy compared to the stand-alone algorithm (93% vs. 87%).Finally, we utilized swarm search graphs to assess the performance of SVM-PSO with ALL-IDB 1&2 in experiment 2.Fig. 17The Swarm search in iteration 8.Fig. 18The Swarm search in iteration 52.Figure 17 presents the hybrid model’s swarm search behavior, resulting in the optimal global solution at the beginning of iteration 8, it shown how the swarm search behave to find the best global solution and how each particle in its best local location. In Fig. 18, the swarm found the best global solution in iteration 52.
