SaveDroid. Public Demo

SaveDroid

Evasion-Aware Android Malware Detection System

SaveDroid is an Android malware detection system built on an optimised K-Nearest Neighbours model designed to stay accurate even when attackers try to evade detection through feature manipulation. The study evaluates the model on a balanced dataset of 36,000 Android apps using hybrid static and dynamic features, and reports up to 99.6% detection accuracy after optimisation, with strong resilience against Feature Injection and Feature Removal attacks.

Read the paper preprint

Run prediction

Labels are optional. If provided, first column must be 0 or 1 and row count must match the features file.
Download template CSV
Input formats
CSV, XLSX, XLS
Expected columns
1433
Evaluation metrics
Labels optional
The uploaded dataset must exactly match the SaveDroid trained feature schema. Files with missing, extra, or reordered columns will be rejected.
Test SaveDroid with a sample dataset

You can download a small sample dataset compatible with the SaveDroid feature structure. This allows you to quickly test the detection system and explore the results dashboard.

Download Sample Features Download Sample Labels

Onyebuchi Sunday Onwuajuese. Edge Hill University. © 2026

Data privacy and usage

SaveDroid operates as a privacy-preserving research demonstration. Uploaded datasets are processed only during the current session in order to generate predictions. The platform does not store, log, or retain uploaded files.

All data is handled in temporary memory and is automatically cleared after the prediction run or when the session expires. Users should avoid uploading sensitive or confidential datasets.