Delivery Predict AI - Neural Network Study Project

A neural network study project for delivery time and price prediction

📋 Overview

Delivery Predict AI is an educational machine learning project developed to explore neural network fundamentals using PyTorch. The project simulates a delivery prediction system similar to existing food delivery platforms like iFood, focusing on predicting delivery time and cost based on distance.

This project serves as a study case for understanding:

• Linear regression with neural networks
• PyTorch fundamentals and training pipelines
• Multi-output prediction models
• Data visualization for ML models

⚠️ Disclaimer: This is an educational project, not a production-ready system. The models and data are simplified for learning purposes.

🎯 Project Objectives

• Study neural network implementation using PyTorch
• Understand linear regression for predictive analytics
• Explore multi-output prediction scenarios
• Create visualizations for model performance analysis
• Practice software architecture for ML projects

🏗️ Project Structure

├── 📁 src
│ ├── 📁 data
│ │ └── 🐍 dataset.py # Sample delivery dataset
│ ├── 📁 model
│ │ └── 🐍 model.py # Neural network architectures
│ ├── 📁 utils
│ │ └── 🐍 graph_helper.py # Visualization utilities
│ └── 🐍 main.py # Main training and evaluation script
├── ⚙️ .gitignore
├── 📄 poetry.lock # Poetry dependency lock file
├── ⚙️ pyproject.toml # Project configuration and dependencies
└── 📝 README.md # This file

🧠 Model Architecture

The project implements two separate linear regression models:

1. Time Prediction Model

• Input: Distance (kilometers)
• Output: Estimated delivery time (minutes)
• Architecture: Single linear layer (nn.Linear(1, 1))

2. Price Prediction Model

• Input: Distance (kilometers)
• Output: Estimated delivery price (Brazilian Reais - R$)
• Architecture: Single linear layer (nn.Linear(1, 1))

Training Parameters:

• Loss Function: Mean Squared Error (MSE)
• Optimizer: Stochastic Gradient Descent (SGD)
• Learning Rate: 0.01
• Epochs: 1000

📊 Dataset

The synthetic dataset simulates delivery metrics for a Honda CG 160 motorcycle:

Distance (km)	Time (minutes)	Price per km (R$/km)
1.0	1.2	0.50
2.0	2.4	1.00
3.0	3.6	1.50
4.0	4.8	2.00
5.0	6.0	2.50
6.0	7.2	3.00
7.0	8.4	3.50
8.0	9.6	4.00
9.0	10.8	4.50
10.0	12.0	5.00

Dataset characteristics for study:

• Linear relationships: Both time and price per km increase linearly with distance
• Time calculation: Based on average speed of 50 km/h → time = (distance / 50) * 60
• Price progression: Price per km increases from R$ 0.50 to R$ 5.00 linearly
• Total price: Can be calculated as total_price = price_per_km × distance

🚀 Getting Started

Prerequisites

• Python 3.8 or higher
• Poetry (dependency management)

Installation

1. Clone the repository:

   git clone https://github.com/yourusername/delivery-predict-ai.git
   cd delivery-predict-ai

2. Install dependencies using Poetry:

   poetry install

3. Activate the virtual environment

   poetry shell

📈 Visualizations

The project generates two main visualizations:

1. delivery_analysis.png - Dual plot showing:

   • Distance vs Time prediction
   • Distance vs Price prediction with R$/km overlay

2. price_per_km_analysis.png - Detailed analysis:

   • Price per kilometer across distance range
   • Comparison against target rate (R$ 2.40/km)

🧪 Example Usage

After training, the model can make predictions:

# Example: Predict for 3.5 km
distance = torch.tensor([[3.5]])
time_pred = time_model(distance)      # Estimated time
price_pred = price_model(distance)    # Estimated price

Dependencies

Core dependencies (managed by Poetry):

• torch>=2.0.0: Deep learning framework
• matplotlib>=3.7.0: Data visualization
• numpy>=1.24.0: Numerical computations

🎓 Learning Outcomes

This project demonstrates:

PyTorch Fundamentals

• Tensor operations and autograd
• Neural network module creation
• Training loop implementation

ML Concepts

• Linear regression as neural networks
• Loss functions and optimization
• Model evaluation metrics

Software Engineering

• Modular project structure
• Code documentation standards
• Visualization integration

🙏 Acknowledgments

• Inspired by real-world delivery platforms like iFood
• Built for educational purposes in neural network studies
• Thanks to the PyTorch community for excellent documentation

Built for learning • Powered by PyTorch • Inspired by real-world applications

_{Note: This project simulates delivery prediction for educational purposes only. Actual delivery platforms use more complex models and real-time data.}