VANET Packet Delivery Prediction

Objective

The goal of this project is to predict whether a data packet in a Vehicular Ad Hoc Network (VANET) will successfully reach its destination. This is a binary classification task based on a dataset of simulated VANET metrics collected in urban scenarios.

The project leverages a deep learning-based neural network model to analyze the following features:

• Available Bandwidth (ABE)
• Trajectory of the vehicle (TJR)
• Number of Neighbors (NV)
• Distance to Destination (DST)
• MAC Layer Losses (LMAC)

Dataset Description

The dataset includes 5823 samples with the following columns:

1. ABE: Available bandwidth in the link.
2. TJR: Trajectory of the vehicle regarding the destination.
3. NV: Number of neighbors.
4. DST: Distance to the destination.
5. LMAC: MAC layer losses.
6. OUT: Target variable (1 for successful delivery, 0 otherwise).

The dataset has been preprocessed and normalized for training and evaluation purposes.

Methodology

1. Data Preprocessing

   • Data is normalized to ensure uniformity across features.
   • The dataset is split into training (80%) and testing (20%) sets.

2. Exploratory Data Analysis (EDA)

   • Correlation heatmap to understand feature relationships.
   • Distribution plots for target variable (OUT).
   • Dataset preview and summary statistics.

3. Model Architecture

   • A Sequential Neural Network with the following layers:
     • Fully connected layers with ReLU activation:
       • 256 -> 128 -> 64 -> 32 units.
     • Output layer with sigmoid activation for binary classification.
   • Optimized using the Adam optimizer with a binary cross-entropy loss function.

4. Evaluation Metrics

   • The model's performance is evaluated using the following metrics:
     • Accuracy = 0.8215
     • Precision = 0.7996
     • Recall = 0.7870
     • F1 Score = 0.7932
   • A confusion matrix is plotted for visual analysis of classification results.

5. Deployment

   • A Streamlit application is built with two tabs:
     • Explore: Displays dataset preview, EDA graphs, and model metrics.
     • Predict: Allows users to input feature values and predict the packet delivery outcome.

How to Run

1. Clone the repository:

   git clone https://github.com/JaiSwarup/vanet-streamlit.git  
   cd vanet-prediction  

2. Install the required dependencies:

   pip install -r requirements.txt  

3. Train the model or use the pre-trained saved model:

   • Training: Run the training script to train the neural network.

     python train_model.py  

   • Using Pre-trained Model: Ensure model.h5 is in the working directory.

4. Run the Streamlit app:

   streamlit run app.py  

Link to the original dataset - https://upcommons.upc.edu/handle/2117/353774