**8. Conclusions**

We have proposed static, dynamic and hybrid methods for detecting malware targeting Android mobile devices. Our three methods are based on fully connected neural networks trained by the Tensorflow/Keras libraries. The static network, reaching an accuracy of 92.9% and a precision of 91.1%, is trained on 840 static features. The dynamic neural network, reaching an accuracy of 81.1% and a precision of 83.4%, is trained on

3722 dynamic features. The hybrid neural network, reaching an accuracy of 91.1% and a precision of 91.0%, is trained on 7081 features (i.e., 3359 statics and 3722 dynamics). Feature selection techniques are used, such as Pearson correlation and a manual method. In addition, we have presented that 22,636 static features and 2210 dynamic features of the Omnidroid dataset are empty for a total of 24,846 out of 31,931 (i.e., 77.81%).

As future work, this research could be generalized to other operating systems, such as iOS, which represents about 20% of the mobile market [38]. At that point, new tools for extracting static and dynamic features should be developed, in order to build a new dataset that we would be labeled by using VirusTotal. In addition, all results related to the learning techniques, the evaluation metrics, as well as the hyperparameter configuration, could be reused for training the neural networks. For further research, it would be necessary to update the dataset with the most recent labelling techniques, and to develop an automation tool for updating neural networks automatically.

**Author Contributions:** Conceptualization, C.R., F.E.K., R.B. and S.P.; methodology, C.R., F.E.K., R.B. and S.P.; software, C.R. and F.E.K.; validation, C.R., F.E.K., R.B. and S.P.; formal analysis, C.R. and F.E.K.; data curation, C.R. and F.E.K.; writing—original draft preparation, F.E.K.; writing—review and editing, F.E.K.; visualization, F.E.K.; supervision, S.P.; project administration, S.P. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research was funded by Natural Sciences and Engineering Research Council of Canada (NSERC), Prompt Quebec, Flex Groups and Q-Links.

**Data Availability Statement:** Publicly available dataset for Omnidroid [5] is analyzed in this study. These data can be found here: (http://aida.etsisi.upm.es/download/omnidroid-dataset-v1 (accessed on 12 September 2021)).

**Acknowledgments:** We would like to express our gratitude to Flex Groups teams for their technical support.

**Conflicts of Interest:** The authors declare no conflict of interest. The funders had no role in the design of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript, or in the decision to publish the results.
