Applied machine learning · Demonstration interface

Historical railway delay prediction

An exploratory study using a saved XGBoost regression model and historical station-delay records. Choose a held-out example to inspect the model's estimate alongside the recorded outcome.

Model evaluation

Chronological test partition · loaded from report

Mean absolute error—Minutes · lower is better

Root mean squared error—Minutes · penalises large errors

R² score—Explained variation · test set

Model familyXGBoostGradient-boosted regression trees

Inspect one test record

01 / PREDICT

Select a held-out example. The prediction button sends its ID to the backend, which runs the saved model.

Historical record

Scheduled date—

Station—

Train category—

Route position—

Uses the selected saved test example

Predicted delay

— minutes

MODEL OUTPUT

Observed delay—

Absolute error—

Choose a record and run the model to compare estimate and observed delay.

Features used by the model

02 / MODEL

Relative importance from the fitted estimator. This describes model behaviour; it does not establish that a feature causes delay.

Feature data will appear when the model service connects.

Bars are scaled relative to the most important feature in this model. Values are not percentages of delay explained.

Data & method

03 / METHOD

Uses historical railway schedule and delay data prepared for this project.
Uses a chronological train/test split to evaluate on later records.
Regression output is a delay estimate in minutes for a selected historical example.
Performance metrics are loaded from the project's saved evaluation report.

Dataset: Kaggle — rxydenxd/indian-railways-delay-dataset