Table 8 Multimodal versus unimodal comparative analysis for CCCS-CIC-AndMal-2020 and Blended malware image datasets.
From: Multimodal malware classification using proposed ensemble deep neural network framework
Ensemble models | Multimodal implementation (Late fusion) | ||||||
|---|---|---|---|---|---|---|---|
RUS-Boost (%) | Random forest (%) | Subspace (%) | AdaBoost-M2 (%) | BagTree (%) | Proposed NNW (Numeric + Visual) (%) | RUSBoost (Numeric) and proposed NNW(Visual) (%) | |
Majority Vote | 94.34 | 94.22 | 55.52 | 79.34 | 92.13 | 94.02 | 95.36 |
Stacked Ensemble | 74.95 | 74.38 | 24.05 | 53.94 | 74.83 | 82.55 | 82.64 |
Boosted Ensemble | 94.34 | 76.13 | 46.84 | 63.11 | 73.83 | 94.02 | 95.04 |
Bagged Ensemble | 94.34 | 76.13 | 50.56 | 64.68 | 73.83 | 94.02 | 95.36 |
Existing Mixed Dataset68 | – | – | – | – | – | 92.3 | – |
Unimodal accuracies | |||||||
|---|---|---|---|---|---|---|---|
Numeric Dataset | 94.2 | 94.42 | 39.70 | 71.93 | 89.04 | 91.40 | 94.2 |
Imagery Dataset | 93.45 | 93.35 | 73.50 | 87.42 | 94.23 | 96.06 | – |
Existing Numeric Dataset69 | - | – | – | – | – | 93.36 | – |
Existing Imagery Dataset70 | – | – | – | – | – | 95 precision | – |
Existing Pure Text68 | – | – | – | – | – | 96.2 | – |
Existing Words-changing Text68 | – | – | – | – | – | 87.3 | – |
Existing Words-missing Text68 | – | – | – | – | – | 92.4 | – |
Existing Pure Image68 | – | – | – | – | – | 91.2 | – |
Existing multimodal (API call + Visual)43 | – | – | – | – | – | – | 93.5 |
