Winners
Congratulations to 2025 Winners!
| First Prize Award Winner($5000) | Sean Williams California, United States |
| Second Prize Award Winner($3000) | Prashanti Rao Seattle, United States |
| Third Prize Award Winner($1000) | Justin Collins, Maine, United States |
2025 US Applied AI Olympiad Challenge: Last-Mile Delay Prediction Challenge
Challenge Overview
In this challenge, you will build a machine learning model that predicts delivery delay time (in minutes) using real-world operational signals such as distance, prep time, traffic, weather, and courier experience.
Your task is to train a model that accurately predicts delays before they happen, enabling smarter dispatch, pricing, and staffing decisions.
This is a pure applied AI problem. Strong feature engineering, modeling choices, and validation strategy matter more than deep theory.
The Problem
Given historical delivery data, predict:
delay_minutes – the total delivery delay beyond the expected time.
Each row represents a single delivery order. The data contains a mix of numerical and categorical features that reflect real operational complexity.
Dataset
You are provided with:
train.csv– labeled training data (features + target)test.csv– unlabeled evaluation data (released later)
Target Variable
delay_minutes(continuous, regression task)
Features (Data Dictionary)
| Feature | Description |
|---|---|
order_id | Unique identifier |
distance_km | Delivery distance |
prep_time_min | Restaurant preparation time |
courier_experience_months | Courier tenure |
weather | Weather condition |
traffic_level | Traffic congestion |
time_of_day | Meal period |
day_of_week | 0=Monday … 6=Sunday |
order_value_usd | Order price |
restaurant_rating | Average rating |
items_count | Number of items |
past_delay_rate | Courier’s historical delay frequency |
surge_multiplier | Dynamic pricing multiplier |
zone | Delivery zone type |
Objective
Build a model that minimizes prediction error on unseen deliveries.
Evaluation Metric
Mean Absolute Error (MAE)
Lower MAE = better performance.
Rules
- Any ML framework allowed (sklearn, XGBoost, LightGBM, CatBoost, PyTorch, etc.)
- External datasets not allowed
- Feature engineering is allowed
- Ensembles are allowed
- No manual labeling or leakage from test data
What We’re Looking For
- Clean data handling
- Thoughtful feature engineering
- Solid validation strategy
- Practical model choices
- Clear reasoning
This challenge mirrors problems faced by logistics companies, delivery platforms, and operations teams in the real world.
Deliverables
Participants must submit the following:
Google Colab Notebook:
Content:
- Preprocessing of the dataset.
- Exploratory data analysis (EDA).
- Implementation of a CNN classifier to predict plant species.
- Evaluation of model performance using metrics like accuracy and confusion matrix.
- Insights drawn from the data and model results.
Submission Format: The notebook must be submitted as an HTML file (.html).
Predictions file
- CSV with columns:
order_id,delay_minutes
Business Presentation:
Audience: Tailor your presentation for a Data Science Lead in the agricultural industry.
Content:
- Business Overview: Clearly define the problem and explain your solution approach.
- Key Findings: Highlight insights derived from the data and model results that drive business decisions.
- Business Recommendations: Provide actionable steps to improve agricultural efficiency using your model.
- Potential Benefits: Explain how the solution can lead to higher yields, reduced manual labor, and sustainable practices.
Submission Format: The presentation must be submitted as a PDF file (.pdf). Avoid copying code into the slides unless necessary to illustrate key points.
Submission Guidelines
- Challenge Opens: Nov 15, 2025
- Submission Deadline: Nov 30, 2025
Submit your deliverables through the competition portal. Ensure that all files are error-free and adhere to the submission formats.
Evaluation Criteria
Submissions will be judged on the following:
- Model Accuracy: How effectively your CNN model classifies the plant species.
- Insights and Analysis: The depth and clarity of insights derived from the data.
- Business Recommendations: Relevance, feasibility, and impact of your proposed actions.
- Presentation Quality: Engaging and clear storytelling tailored to a business audience.
Best Practices for Success
- Use Google Colab’s GPU runtime for faster training.
- Document your notebook with inline comments and markdown cells for better readability.
- Ensure the notebook runs sequentially without warnings or errors.
- Keep your presentation concise, with a focus on business impact and actionable insights.
Why Participate?
- Gain hands-on experience with real-world datasets and deep learning techniques.
- Develop and showcase your data science and business communication skills.
- Be part of the growing movement to integrate AI into agriculture and solve global challenges.
Join us in transforming agriculture with AI—your innovations today can shape a sustainable tomorrow!
