SDG 9 — Industry & Infrastructure SDG 12 — Responsible Consumption Binary Classification · Decision Tree

Predicting Late Delivery Risk in Construction Supply Chains

An AI-powered classifier that predicts whether a shipment will be delayed — enabling procurement teams to act before disruptions stall critical infrastructure projects.

Try Live Predictor → View ML Pipeline
81.93%
Accuracy
0.82
ROC AUC
177K
Training Records
29
Features
v.3
Deployed Model
Component 2

Data Analysis & Dashboard

Exploratory analysis of 177,519 shipment records from the DataCoSupplyChain dataset — uncovering the patterns that drive late delivery risk across global construction supply chains.

01
Scheduled Days is the Strongest Predictor
Shipments with a scheduled duration of 4+ days show dramatically higher late-delivery rates. Planning assumptions built into procurement contracts are routinely optimistic, creating a systemic risk gap.
02
Geography Creates Distinct Risk Tiers
Markets in Southeast Asia, West Africa, and Central America show consistently higher delay rates — reflecting real infrastructure bottlenecks, port congestion, and road network limitations that historical models must account for.
03
Profit Erosion Tracks Directly with Delays
The Order_Item_Profit_Ratio averages 0.12 overall, but drops to -2.75 at its minimum — values correlated with late shipments. Late deliveries are not just operational failures; they are profit destruction events.
04
Shipping Mode is a Key Risk Lever
Standard Class shipping is the dominant mode and also the highest-risk. Upgrading at-risk shipments to First Class or Same Day is an actionable, data-supported mitigation strategy for high-priority orders.

Feature Distributions (EDA Histograms)

Benefit per order distribution
Benefit per Order
Sales distribution
Sales per Customer
Item discount distribution
Order Item Discount
Product price distribution
Product Price
Profit ratio distribution
Item Profit Ratio
Order item total distribution
Order Item Total
Order profit distribution
Order Profit per Order
Product price 2 distribution
Product Price (Raw)
📊
Power BI / GenAI Dashboard
Interactive dashboard with full EDA, trends, and stakeholder visuals.
Dashboard link — coming soon
Component 3

ML Pipeline

End-to-end workflow built on the Braintoy MLOS platform — from raw data ingestion through model governance and live deployment.

1

Dataset Upload & Target Definition

Uploaded the DataCoSupplyChain CSV (177,519 records, 29 columns) into MLOS. Locked Late_delivery_risk as the binary target variable (1 = Late, 0 = On Time) with an 80/20 train/validation split.

MLOS dataset upload and target selection
2

Categorical Feature Encoding

Applied Categorical-to-Numeric transformation to 11 high-cardinality features including Type, Category_Name, Shipping_Mode, Market, and Customer_Segment — enabling decision tree architectures to process complex logistical labels.

MLOS feature encoding step
3

Model Training — Comparative Evaluation

Trained two candidate models: DecisionTreeClassifier (v.3) and ExtraTreesClassifier (v.1). Both were evaluated on Accuracy, F1-Score, ROC AUC, Precision/Recall, and Hamming Loss.

Model training run 1 Model training run 2
4

Model Evaluation & Selection

DecisionTreeClassifier (v.3) selected as the deployment candidate with 81.93% accuracy and 0.82 ROC AUC — outperforming ExtraTreesClassifier on accuracy and Hamming Loss (0.18 vs. 0.24).

81.93%
Accuracy
81.93%
F1-Score
0.82
ROC AUC
0.18
Hamming Loss
Confusion Matrix
Confusion Matrix — Model v.3 (DecisionTree)
ROC Curve
ROC Curve — AUC 0.82
Precision vs Recall Curve
Precision vs. Recall Curve
Feature Importance Chart
Feature Importance — Top Predictors
5

Governance Review & Approval

Model container submitted to fariha@braintoy.ai for third-party ethical validation per BuiltSmart AI governance requirements — verifying the model's logic is mathematically stable and free of critical biases before deployment.

MLOS governance submission
6

Deployment — Live MLOS Web App

CapstoneML v.3 deployed on the Braintoy MLOS platform as a no-code interactive interface. Non-technical users — site managers, procurement officers — can input shipment parameters and receive an instant Late/On-Time risk label.

MLOS deployment interface MLOS prediction output
Component 4

Explore Prediction

Enter real shipment parameters below and the trained CapstoneML v.3 model will predict whether that order is at risk of being delivered late. You will need your MLOS API key and access token.

Output Explained

1 — Late Delivery Risk
High probability the shipment will arrive after the scheduled date. Recommend upgrading shipping mode or flagging for manual review.
0 — On Time
Model predicts delivery will arrive on or before the scheduled date based on historical patterns.

Key Input Drivers

  • 🚚 Shipping Mode — Standard vs. First Class
  • 📅 Scheduled Days — Planned lead time
  • 🌍 Market / Region — Geographic risk tier
  • 💰 Profit Ratio — Financial exposure
  • 📦 Order Quantity — Volume risk

CapstoneML v.3 — Live Prediction Interface

Live Model
⚠️ This prediction is powered by your trained MLOS model. You must enter your API Key and Access Token from your Braintoy MLOS account to activate predictions. Results are AI-generated and should be reviewed by a qualified logistics professional before acting on them.
Component 8 — Resources

Project Resources

All deliverables for the BuiltSmart AI Capstone — BuiltSmart AI Cohort, September 2026.

📊
Dataset
DataCoSupplyChain — Kaggle
📄
Executive Report
Full written report PDF
🎞️
Presentation Slides
Final slide deck
Coming Soon
🤖
AI Chatbot
JotForm AI Agent — Ask the model
🎬
NotebookLM Video
Audio/video summary
📈
Power BI Dashboard
Interactive data dashboard
Coming Soon
Component 5 — AI Chatbot

Ask the AI Agent

Chat with the JotForm AI Agent trained on this capstone project. Ask it about late delivery risk, the dataset, model performance, or supply chain strategy.

AI Agent — Online

Your capstone expert is ready to chat

The JotForm AI Agent has been trained on this project's dataset, methodology, and findings. It can answer questions about the model, explain predictions, or discuss supply chain strategy — available 24/7 via the chat button.

Look for the chat bubble in the bottom-right corner of this page.

Try asking…

Powered by JotForm AI Agent  ·  Responses are AI-generated and for informational purposes only.

About

About Me

BuiltSmart AI Cohort Machine Learning Supply Chain SDG 9 & 12
Francis Aoo
BuiltSmart AI Cohort — Applied Machine Learning

This capstone represents the culmination of the BuiltSmart AI program — applying machine learning to a domain I care about deeply: the reliability of construction supply chains. Late deliveries don't just cause logistical headaches; they stall infrastructure projects that communities depend on, erode contractor margins, and undermine the sustainability commitments we need to meet global SDG targets.

By building CapstoneML v.3, I wanted to demonstrate that AI doesn't need to be complex to be useful. A well-trained classifier on real shipping data can give procurement teams a practical early-warning signal — transforming logistics from reactive firefighting to intelligent prevention.

🌐 Visit Vouchya.com