Delivery Time Prediction Project

This is a machine learning project to predict delivery times based on various features like traffic, weather etc. It follows a modular structure for easy maintenance.

1. Firstly Create your new Environment

conda create -p env python=3.9 -y

2. Activate Conda Environment

conda activate env/ -> CMD

source activate env/ -> Git bash

3. Install necessary requirements

pip install -r requirements.txt

Project Structure

- src
  - components
    - data_ingestion.py
    - data_transformation.py 
    - model_trainer.py
  - config
    - configuration.py
  - constants
    - constants.py
  - exception
    - exception.py
  - logger
    - logger.py
  - pipeline
    - training_pipeline.py
  - utils
    - __init__.py
- artifacts
- data
- logs
- notebooks
- README.md

Usage

The main entry point is training_pipeline.py. It does the following:

1. Calls data_ingestion.py to split the raw data into train and test sets.
2. Calls data_transformation.py to preprocess the data and engineer features.
3. Calls model_trainer.py to train and evaluate models. It saves the best model.
4. Configuration parameters are stored in config/configuration.py.
5. Logger and custom exceptions are handled in logger.py and exception.py.
6. Common utility functions are kept in utils.py.

The output model and intermediate artifacts are saved in the artifacts folder.

How to run Prediction

python src/components/data_ingestion.py

This will do all the steps and generate the model file in artifacts/model_trainer.

How to make predictions

python prediction/batch.py

Dataset

The dataset used for this project is data/finalTrain.csv. It contains the following features:

• ID: Unique ID for each row
• Type_of_order: Online or offline order
• Type_of_vehicle: Bike, Car etc.
• Time_taken: Target variable, delivery time in minutes.

Data Ingestion

The data_ingestion.py does the following:

• Load data from csv file
• Split into train and test sets
• Save train and test csv files in the artifacts/data_ingestion folder.

Data Transformation

The main steps done in data_transformation.py:

• Handle missing values
• Encode categorical features
• Standardize numerical columns
• Engineer new features like distance calculation
• Save the transformed data in artifacts/data_transformation folder.

It also saves the preprocessing object for use during prediction.

Model Training

model_trainer.py trains the following regression models:

• XGBRegressor
• RandomForestRegressor
• SVM

It evaluates them based on R2 score and saves the best model to artifacts/model_trainer.

Deployment

The model can be deployed using Flask by creating a app.py file that loads the model and serves predictions via an API.

Monitoring

The model's performance can be monitored in production by tracking live metrics like request throughput, latency, error rates etc.

Notes

• This project demonstrates how to structure a ML project with separation of concerns between components.
• Logging and exception handling is done using custom logger and exception classes.
• Ideal for learning how modular ML projects are built.
• Can be extended to include more components like model evaluation, deployment etc.