Table 4 Experiments on Datafountain BDCI-21 Dataset.
From: Semantic lossless encoded image representation for malware classification
Model | Year | Techniques | Normalized image size | Acc | \(F1_{macro}\) | \(FPR_{macro}\) |
|---|---|---|---|---|---|---|
KNN | – | – | \(512\times 512\) | 0.969 | 0.964 | 0.008 |
DT | – | – | \(512\times 512\) | 0.955 | 0.954 | 0.01 |
XGBoost | – | – | \(512\times 512\) | 0.977 | 0.972 | 0.004 |
AlGarni et al.58 | 2022 | EfficientNet | \(224\times 224\) | 0.987 | 0.987 | 0.003 |
Parihar et al.59 | 2022 | ResNet+Xception+EfficientNet | \(224\times 224\) | 0.990 | 0.990 | 0.001 |
Kumar et al.21 | 2022 | Deep CNN(VGG16 based) | \(224\times 224\) | 0.992 | 0.991 | 0.001 |
Son et al.18 | 2022 | k-NN+SVM+CNN | \(32\times 64\) | 0.971 | 0.971 | 0.004 |
Mallik et al.20 | 2022 | VGG16+BiLSTM | \(224\times 224\) | 0.982 | 0.980 | 0.002 |
Rustam et al.22 | 2023 | VVG16+ResNet | \(224\times 224\) | 0.989 | 0.989 | 0.002 |
Proposed method | – | CNN+SPP+Transformer | No Limit | 0.997 | 0.997 | 0.001 |
