Skip to content

Latest commit

 

History

History
188 lines (163 loc) · 7.83 KB

README.md

File metadata and controls

188 lines (163 loc) · 7.83 KB
Raw

This repository includes the code for the published paper CNN-Based Continuous Authentication of Smartphones Using Mobile Sensors. It features the model architecture, data collection tools, and methods for feature extraction.
Paper Link

Installation instructions

  1. Create a new conda environment conda create -n userauth python=3.8
  2. Activate the environment conda activate userauth
  3. Install the required dependencies pip install -r requirements.txt

Prepare training dataset for REST filter model

cd prepare_dataset

You can directly download our dataset rest_data or Create your dataset with our app.
image

Note: accel.csv and gyro.csv is store at Android/data/com.example.user_auth/files
Rename accel.csv -> Accel.csv and gyro.csv->Gyro.csv.
a. Create dataset with app:
Collect REST data:
  Place your phone in table and click Start to collect the data untill it reaches 5MB and change the position
(1) Upright, 2) Downface, 3) Upface) make REST folder maintain the metioned directory structure

Collect MOTION data:
  Start the app and perform following activities
 1. scrolling
 2. talking in phone
 3. chatting
 4. Walking
 5. Jumping etc.

Directory structure:
dataset
    └── rest_data
        ├── raw
        │   ├── test
        │   │   ├── MOTION
        │   │   │   ├── 1
        │   │   │   │   ├── Accel.csv
        │   │   │   │   └── Gyro.csv
        │   │   │   ├── 2
        │   │   │   │   ├── Accel.csv
        │   │   │   │   └── Gyro.csv
        │   │   │   └── 3
        │   │   │       ├── Accel.csv
        │   │   │       └── Gyro.csv
        │   │   └── REST
        │   │       ├── 1
        │   │       │   ├── Accel.csv
        │   │       │   └── Gyro.csv
        │   │       ├── 2
        │   │       │   ├── Accel.csv
        │   │       │   └── Gyro.csv
        │   │       └── 3
        │   │           ├── Accel.csv
        │   │           └── Gyro.csv
        │   └── train
        │       ├── MOTION
        │       │   ├── 1
        │       │   │   ├── Accel.csv
        │       │   │   └── Gyro.csv
        │       │   ├── 2
        │       │   │   ├── Accel.csv
        │       │   │   └── Gyro.csv
        │       │   ├── 3
        │       │   │   ├── Accel.csv
        │       │   │   └── Gyro.csv
        │       │   └── 4
        │       │       ├── Accel.csv
        │       │       └── Gyro.csv
        │       └── REST
        │           ├── 1
        │           │   ├── Accel.csv
        │           │   └── Gyro.csv
        │           ├── 2
        │           │   ├── Accel.csv
        │           │   └── Gyro.csv
        │           ├── 3
        │           │   ├── Accel.csv
        │           │   └── Gyro.csv
        │           └── 4
        │               ├── Accel.csv
        │               └── Gyro.csv

Run following code to prepare rest filter training dataset:
This code creates dataset/rest_data/train folder.

python  prepare_rest_filter_dataset.py

Train REST filter model

You can directly download the models: Rest

Model directory structure
weights
└── Rest
    ├── Features
    │   ├── coorel_features.txt
    │   ├── selector.joblib
    │   └── std_scaler.joblib
    └── Model
        └── model.sav

To train the model, run the following command.

cd ..
python train_restfilter.py

This command creates model.sav at weights/Rest/Model/ and pre-processing rest features
coorel_features.txt, selector.joblib and std_scaler.joblib at location weights/Rest/Features

The model.sav , coorel_features.txt, selector.joblib and std_scaler.joblib are used to filter the rest or motion data during CNN data perparation and testing.

Prepare training dataset for CNN model:

You can directly download our dataset raw_data, data_processed and cnn_training_dataset for testing and can assign as intruder

Create dataset with app:
  Start the app, it will automatically collects the data when you are active.
  copy the Accel.csv and Gyro.csv in the following directory structure

Directory structure:
dataset
└── raw_data
    ├── 37905f086c46d416
    │   ├── Accel.csv
    │   └── Gyro.csv
    ├── 79435f784571ade9
    │   ├── Accel.csv
    │   └── Gyro.csv
    ├── 9b09051ccc1f3dc4
    │   ├── Accel.csv
    │   └── Gyro.csv
    └── f66e251ee7c7dbae
        ├── Accel.csv
        └── Gyro.csv

Run the following code to prepare CNN training dataset

cd prepare_dataset
python prepare_cnn_dataset.py --id 'your_id_name' --pre_process_raw_data

It takes 1 - 2 hour according to the size of dataset.

This code creates data_processed and cnn_training_dataset folder at dataset
where data_processed contains pre-processed data for each users and the cnn_training_dataset
contains the training dataset for the user (--id 'your_id_name') and all the users are intruder for
'your_id_name'.

Once the raw data is pre-processed for all the available users. You don't need to pass --pre_process_raw_data.
you can directly use following code

cd prepare_dataset
python prepare_cnn_dataset.py --id 'your_id_name'
  1. For CNN model

Train the model

To train the model, run the following command.

    python train.py --id user_id --kernel_size kernel_size --window_size window_size --lr learning_rate --num_filters number_of_filters_to_use --epoch number_of_epochs_to_train --dataset_path root_path_of_the_dataset

Example:

    python train.py --id 37905f086c46d416 --epoch 50

Test your model

You can prepare testing dataset with the above method To run the testing code:

python test.py --id 'your_id' --X 'dataset path' --Y 'dataset label path' --model_path 'Trained model checkpoint path (.h5)'

Example:

python test.py --id 37905f086c46d416 --X ./dataset/cnn_training_dataset/37905f086c46d416/X_test.csv --Y ./dataset/cnn_training_dataset/37905f086c46d416/Y_test.csv --model_path ./weights/37905f086c46d416/Models/cp.h5