What is truck-delay-prediction?

This repository contains a machine learning project that predicts whether a truck shipment will be delayed. Given details about a trip, such as the distance, truck age, driver experience, cargo weight, weather conditions, road quality, and number of stops, the system outputs a prediction of delayed or on time along with a confidence percentage and a risk level. The project is structured as a pipeline with several stages. It pulls data from MySQL and PostgreSQL databases, cleans and transforms the data into features the model can use, trains three different prediction models, and compares their accuracy. The three models tested are Random Forest, XGBoost, and LightGBM. According to the README, LightGBM performed best and is used in production. A mock data mode is included so the entire pipeline can be run without any real database connection. Once trained, the model is served through a web API built with Flask. The API accepts a single shipment's details and returns a prediction, or it can process up to one thousand predictions at once in a batch request. A separate endpoint lets you reload the model without restarting the server. All predictions are logged to a file for monitoring. The project uses MLflow to track each training run, which lets you compare model performance across experiments visually. Configuration is handled through a single YAML file and environment variables so no values are hardcoded into the source files. Unit tests cover the main pipeline stages. Setup requires Python, cloning the repository, installing dependencies, and either providing database credentials or running in mock mode. The README is written for deployment on a platform called Lightning.ai.