|
Source | Approach and year | Name/Year of dataset | Time frame | Features and approach | Dynamic | Malware detection | Category classification |
|
[42] | Semisupervised learning (2018) | 2016 | N/A | Hybrid feature selection: aggregate information | Emulator | Accuracy: 91.23% | N/A (dynamic accuracy: 80.3%) |
[38] | Deep artificial Neural network (2021) | CICANDMAL2019 | N/A | Hybrid feature selection: N/A | Real device | Static accuracy: 93.40% | 4 categories (static accuracy: 92.5%) (dynamic accuracy: 80.3%) |
[43] | Machine learning (2018) | Updroid | 2014–2018 | Hybrid feature selection: N/A | Emulator | Detection as categorization | Accuracy: 96.37% |
[44] | Deep learning (2020) | DL-driod dataset 2019 | N/A | Hybrid Feature ranking: InfoGain | Real device | Accuracy: 98.5% | N/A |
[18] | Deep neural network with pseudolabel (2020) | CICMALDroid2020 | 2017-2018 | Dynamic feature selection: N/A | Emulator | Detection as categorization | 4 categories (F1-score: 97.8%) |
[7] | Machine learning (2021) | CCS-CICAndMal2020 | N/A | Dynamic feature selection: N/A | Emulator | Detection as categorization | 12 categories (precision: 98.4%) |
[45] | Machine learning (2017) | Drebin,Mcafee Praguard | N/A | Static feature selection: mean decrease impurity (MDI) | N/A | 99.82% | 99.26% into families |
[46] | Machine learning (2018) | Drebin, Genome VirusShare | 2009–2017 | Dynamic feature selection: N/A | Emulator | 97.4% | 97.8% into families |
[47] | Machine learning (2020) | VirusShare AndroZoo | 2010–2017 | Dynamic feature importance: Top100 | Emulator | F1: 92.88% (same year) F1: 71.81% (across year) | N/A |
[48] | Machine learning (2021) | APKPure, random dataset | N/A | Static, feature selection: LR based | N/A | Accuracy: 96.3% | N/A |
[49] | Machine learning (2021) | APKPure, VirusShare | N/A | Static, feature selection: filter-based | N/A | F-measure: 95% | N/A |
[50] | Machine learning (2022) | Mendeley repository | N/A | Dynamic, features selection: embedded BFE | Emulator | F-measure: 99% | N/A |
[51] | Machine learning (2022) | Multiple repositories | N/A | Dynamic, feature selection: rough set analysis (RSA) and principal component analysis (PCA) | Emulator | Detection rate: 98.8% | N/A |
[52] | Machine learning (2023) | AndroZoo and Drebin | N/A | Static, feature selection: wrapper based (DDQN) | N/A | Detection rate: 95.6% | N/A |
[23] | Deep neural network with pseudo label stack auto encoder (2022) | CICMALDroid2020 | 2017-2018 | Hybrid feature selection: N/A | Emulator | Accuracy: 98.28% | 5 categories (precision: 98.4%) |
[53] | Ensemble (random forest) | KronoDroid | 2008–2020 | Hybrid, features selection: Chi-squared | Real and emulated device | Accuracy: 95% Precision: 95% | N/A |
Our | Machine learning (random forest) | Subset of 2020 KronoDroid | 2008–2020 | Hybrid, filter-based features selection | Real device | Accuracy: 98.03% | 15 categories (accuracy: 87.6%) |
|