**11. Results**

We present our ideas to each Deep Transfer Learning using a confusion matrix, with each picture explaining a distinct form of malware. As we can see from the list


### **Table 1.**

*List of rates computed from a confusion matrix.*

**Figure 3.** *Confusion matrix for mobileNetV2.*

of rates in **Table 1**, we can determine that the best system for predicting our virus is MobileNetV2, which is the greatest pick from the list of rates.

In **Figure 3**, we can see from the confusion matrix that MobileNetV2 is the best model since it has a significant number of real values on the left and predicted values on the right. The proper productions will occur on the matrix's diagonal.

We have an emphasis on precision or specialization in our work, and these matrices can explain erroneous negatives. We require false negatives matrices for nonmalware detected by malware filters. In other words, positive class malware is malware, while false negatives are not malware. In this case, false negatives are preferable to false positives.

**Table 2** shows the training, validation, and testing trials we conducted for each model. At this point, we may infer that one model detects malware far better than another. As the analysis from the confusion matrix shows, the best one is obviously MobileNetV2.

The ROC curve in **Figure 4** shows that it can accurately detect the kind of malware. The capacity to appropriately analyze and recognize a picture from another side whether it was typical.


### **Table 2.**

*Data accuracy after transfer learning—Performance metrics.*

**Figure 4.** *ROC curve for mobileNetV2.*
