|
|
--- |
|
|
license: apache-2.0 |
|
|
language: en |
|
|
library_name: keras |
|
|
tags: |
|
|
- intrusion-detection |
|
|
- cyber-physical-systems |
|
|
- iot-security |
|
|
- lstm |
|
|
- time-series |
|
|
- cybersecurity |
|
|
datasets: |
|
|
- ToN_IoT |
|
|
--- |
|
|
|
|
|
# ClimIDS: Sensor-Layer Intrusion Detection System |
|
|
|
|
|
This model card is for **ClimIDS**, a lightweight, LSTM-based intrusion detection system (IDS) for the physical sensor layer of IoT deployments. |
|
|
|
|
|
## Model Description |
|
|
ClimIDS analyzes time-series data from environmental sensors (temperature, pressure, humidity) to detect anomalies in climate-monitoring systems. Its lightweight architecture (~5,000 parameters) makes it suitable for edge devices. |
|
|
|
|
|
- **Architecture:** `LSTM -> Dropout -> Dense -> Dense (Sigmoid)` |
|
|
- **Dataset:** Trained on `IoT_Weather` subset of ToN_IoT |
|
|
- **Performance:** 98.81% accuracy, 99.7% attack recall |
|
|
|
|
|
## Intended Use |
|
|
- **Primary Use:** Real-time binary classification of sensor telemetry |
|
|
- **Input:** `(batch_size, 10, 3)` — features `[temperature, pressure, humidity]`, normalized |
|
|
- **Output:** Float between 0.0 (Normal) and 1.0 (Attack), threshold 0.5 |
|
|
|
|
|
## How to Use |
|
|
```python |
|
|
import tensorflow as tf |
|
|
import numpy as np |
|
|
from huggingface_hub import hf_hub_download |
|
|
|
|
|
MODEL_PATH = hf_hub_download("Codelord01/sensor_binary", "sensor_binary.keras") |
|
|
model = tf.keras.models.load_model(MODEL_PATH) |
|
|
model.summary() |
|
|
|
|
|
sample_data = np.random.rand(1, 10, 3).astype(np.float32) |
|
|
prediction_prob = model.predict(sample_data) |
|
|
predicted_class = 1 if prediction_prob > 0.5 else 0 |
|
|
print(f"Prediction Probability: {prediction_prob:.4f}") |
|
|
print("Anomaly Detected" if predicted_class == 1 else "Normal Conditions") |
|
|
|
