A Machine Learning-Powered Real-Time Hardware Profiler and Security Sentinel for Android.
Standard Android battery monitors rely on OS-level statistics to tell you what drained your battery after the fact. Watt-Watch takes a different approach. It acts as an active security sentinel, reading raw hardware telemetry in real-time to detect abnormal power consumption—such as hidden crypto-miners, spyware, or rogue background processes—the exact second it happens.
- Raw Hardware Telemetry: Bypasses the Android API to read directly from
/sys/class/and/sys/kernel/nodes, tracking CPU frequency, GPU load, and Screen Brightness. - Dual-Radio Analysis: Explicitly separates Wi-Fi (
wlan0) and Cellular (ccmni1) traffic to accurately calculate the differing physical power costs of radio transmissions. - Edge AI via Model Transpilation: Trained using XGBoost on a PC, the AI model is transpiled into pure, dependency-free Python using
m2cgen. This bypasses Termux/Android SELinux restrictions and missing ARM-compiled libraries, allowing complex machine learning inference to run natively and efficiently on edge devices. - Real-Time Sentinel: Runs quietly in the background via Termux. If the actual physical battery current spikes significantly higher than the AI's predicted current, it triggers an instant system alert.
Figure 2: XGBoost Feature Importance revealing the specific power personality of the device.
- Logging Phase (
watt_logger.py): Collects a baseline dataset of hardware metrics across various states (idle, gaming, downloading). - Training Phase (PC): An XGBoost regressor is trained on the CSV data to understand the standard energy footprint of the specific phone.
- Transpilation (
translate.py): The heavy XGBoost model is converted into raw, pure Python math logic (brain.py). - Monitoring Phase (
checker.py): The lightweight sentinel watches the hardware nodes 24/7 and flags power anomalies instantly.
Prerequisites: An Android device with Termux installed. No root required for basic node access (depending on OEM).
- Clone the repository:
git clone [https://github.com/sowmiksudo/WattWatch.git](https://github.com/sowmiksudo/WattWatch.git) cd WattWatch
2. **Run the Logger (Create your baseline):**
```bash
python watt_logger.py
Run this for about 20 minutes while performing different tasks to generate a rich watt_watch_data.csv.
3. Train & Transpile (On PC):
Transfer the CSV to your PC and run the training script. This will generate the pure-Python brain.py file.
python train_v2.py
python translate.py
- Deploy the Sentinel (On Phone):
Transfer
brain.pyback to your Termux environment and start the security tool:
python checker.py
I am passionate about the intersection of Data Science, cybersecurity, and low-level system architecture. I enjoy building tools that extract meaning from the noise of raw hardware and network data.
If you find this project interesting, be sure to check out my other cybersecurity tooling:
- TCP Traffic Analyzer for Android: A dynamic analysis tool built using Frida and Python to intercept, inspect, and reverse-engineer raw TCP traffic from Android applications.
Currently preparing for a Bachelor's degree in Data Science and Analytics, exploring edge AI and anomaly detection.
Distributed under the MIT License. See LICENSE for more information.