📦 Opsium Project

Explainable Demand Forecasting & Risk-Aware Capacity Optimization for FedEx Tricolor Network

End-to-end explainable framework that converts demand signals into operationally trusted flight capacity decisions

📌 Overview

This project delivers an end-to-end, explainable framework that converts demand signals into operationally trusted flight capacity decisions for the FedEx Tricolor Challenge. Instead of optimizing demand forecasts in isolation, the solution focuses on bridging the gap between planning and execution—a key challenge in large logistics networks such as FedEx.

The framework ensures that demand forecasts are treated as inputs, not capacity targets, and that capacity allocation reflects cost exposure, delay risk, and operational flexibility. The solution prioritizes interpretability, realism, and decision scalability over black-box accuracy.

🎯 Problem Statement

Traditional forecast-driven planning often leads to:

• White-tail capacity from over-commitment
• Load factor mismatch between planned and actual utilization
• Increased delay risk on fragile routes
• One-size-fits-all utilization targets

Objective

Design a framework that converts demand forecasts into risk-aware, cost-aware capacity decisions, reducing planning-execution gaps without adding flights.

🧠 Solution Overview

The solution is structured into two integrated segments:

Segment 1 — Explainable Demand Forecasting (Round-02)

• Signal-aware, regression-based forecasting
• Demand decomposed into:
  • Base demand
  • Promotion-driven volatility
  • Sentiment and sustainability alignment
• Forecasts include confidence and stability indicators

Segment 2 — Capacity Decision Framework (Round-03)

• Rule-based, explainable capacity logic (no black-box optimization)
• Capacity decisions governed by a 4-Factor Decision Lens:
  • Demand Stability
  • Cost Exposure (Fixed vs Variable)
  • Delay Risk
  • Real-Time Flexibility

🔑 Key Insight

Two routes with identical demand forecasts may require opposite capacity decisions.

🔁 End-to-End Flow

Customer & External Signals
(Promotions • Sentiment • Sustainability • Regulation)
                    ↓
Explainable Demand Forecasting (Segment 1)
(SKU- & Route-Level Forecast + Confidence)
                    ↓
Forecasted Demand as Planning Input
                    ↓
Capacity Decision Framework (Segment 2)
(Cost • Risk • Flexibility Lens)
                    ↓
Operational Decisions
(Utilization • Buffering • Conservative Loading)

📊 Synthetic Dataset Design

Why Synthetic Data?

No official dataset was provided. To ensure realism and explainability, domain-informed synthetic datasets were created to reflect real-world logistics behavior while enabling controlled experimentation.

Dataset 1 — Customer SKU Demand Signals

File: customer_sku_demand_signals.csv
Purpose: Capture customer-level demand drivers and volatility

Key Features:

• base_demand — organic demand baseline
• promotion_flag, discount_percentage — short-term demand shocks
• sentiment_score, review_volume — digital perception signals
• sustainable_sku_flag, eco_preference_index — long-term demand stability
• regulation_impact_score — regulatory influence

Dataset 2 — Forecasted Demand Output

File: forecasted_demand_output.csv
Purpose: Store model-generated, time-bound demand forecasts used directly for capacity planning

Outputs:

• forecasted_demand
• forecast_confidence

Dataset 3 — Tricolor Flight Capacity Metadata

File: tricolor_flight_capacity.csv
Purpose: Represent FedEx-like operational constraints

Key Features:

• max_capacity
• fixed_cost
• variable_cost_per_unit
• real_time_update_flag
• delay_risk_score

Dataset 4 — Round-03 Capacity Decisions (FINAL OUTPUT)

File: round3_capacity_decisions.csv
Purpose: Contains route-level capacity strategy decisions generated using the 4-Factor Decision Lens

Key Columns:

• route
• forecasted_demand
• max_capacity
• demand_stability
• cost_profile
• delay_risk_flag
• flexibility_flag
• utilization_strategy
• planning_comment

This file represents the final operational output of the project.

⚙️ Modeling & Decision Logic

Demand Forecasting (Segment 1)

• Model: Regression-based, interpretable
• Goal: Explain why demand changes, not just how much

Capacity Decision Logic (Segment 2)

Route Characteristics	Capacity Strategy
Stable demand + High fixed cost	Maximize utilization
Volatile demand + Real-time flexibility	Dynamic buffer
High delay risk	Conservative loading
Low cost + Flexible operations	Absorb uncertainty

We do not change demand forecasts — we change how much we trust them operationally.

📈 Business Impact

• ✅ Reduced white-tail capacity through targeted utilization
• ✅ Improved alignment between planned and actual load factors
• ✅ Lower delay risk on fragile routes
• ✅ Explicit trade-offs between cost, risk, and reliability
• ✅ Scalable framework applicable across the Tricolor network

Executive Insight

For FedEx, the cost of being wrong on a high-risk route is higher than flying slightly empty — and this framework makes that trade-off explicit.

📁 Project Structure

Opsium/
├── Opsium_Notebook.ipynb              # End-to-end notebook (Round-02 + Round-03)
├── data/
│   ├── customer_sku_demand_signals.csv
│   ├── forecasted_demand_output.csv
│   ├── tricolor_flight_capacity.csv
│   └── round3_capacity_decisions.csv  # Final Round-03 output
├── notebooks/                         # Optional supporting notebooks
└── README.md

▶️ Usage

1. Open Opsium_Notebook.ipynb in Jupyter or VS Code
2. Ensure all CSV files are present in the data/ directory
3. Run cells sequentially to reproduce:
   • Demand forecasting
   • Capacity classification
   • Strategy visualizations

📦 Dependencies

• Python 3.x
• Pandas
• NumPy
• Scikit-learn
• Matplotlib

Install via:

pip install pandas numpy scikit-learn matplotlib

🚀 Quick Start

Prerequisites

• Python 3.8 or higher
• Jupyter Notebook or JupyterLab

Installation

1. Clone the repository

   git clone https://github.com/Partha0003/Opsium.git
   cd Opsium

2. Install dependencies

   pip install pandas numpy scikit-learn matplotlib

3. Launch Jupyter Notebook

   jupyter notebook Opsium_Notebook.ipynb

🔬 Technical Implementation

Methodology

1. Demand Analysis

• Forecast Validation: Confidence interval analysis
• Trend Identification: Seasonal and promotional patterns
• Risk Assessment: Demand uncertainty quantification

2. Capacity Optimization

• Multi-Objective Framework: Cost vs. service level balance
• Constraint Handling: Operational and regulatory limitations
• Scenario Planning: What-if analysis for different demand scenarios

3. Decision Framework

• Rule-Based Logic: Explainable decision trees
• 4-Factor Decision Lens: Systematic capacity evaluation
• Hybrid Approach: Combined analytical and heuristic methods

Key Performance Indicators

Operational Metrics

• Capacity Utilization: Percentage of available capacity used
• Cost Efficiency: Cost per unit transported
• Service Level: On-time delivery performance
• Forecast Accuracy: Demand prediction precision

Optimization Metrics

• Total Cost Minimization: Overall operational cost reduction
• Risk-Adjusted Returns: Performance considering uncertainty
• Resource Allocation Efficiency: Optimal capacity distribution

🛠️ Advanced Features

Risk Management Components

• Demand Uncertainty: Confidence-based capacity buffering
• Capacity Constraints: Dynamic constraint handling
• Cost Volatility: Multi-scenario cost optimization

🎯 Results & Performance

Framework Achievements

• 🎯 Explainable Decision Making: Every capacity decision includes reasoning
• 📈 Operational Alignment: Reduced planning-execution gaps
• ⚡ Scalable Implementation: Framework applicable across route networks
• 🎪 Risk-Aware Planning: Explicit trade-offs between cost and reliability

Key Insights

1. Demand Patterns: Stability matters more than absolute volume for capacity decisions
2. Cost Optimization: Fixed vs. variable cost structure drives utilization strategy
3. Risk Factors: High-risk routes require conservative capacity allocation
4. Operational Efficiency: Rule-based decisions outperform black-box optimization for explainability

🔍 Notebook Implementation

The comprehensive notebook covers both segments of the solution:

Segment 1: Explainable Demand Forecasting

1. Data Loading & Signal Processing

   • Import customer SKU demand signals
   • Feature engineering for demand drivers
   • Signal validation and preprocessing

2. Demand Decomposition

   • Base demand calculation
   • Promotion impact analysis
   • Sentiment and sustainability factors
   • Regulatory influence assessment

3. Forecast Generation

   • Regression-based demand prediction
   • Confidence interval calculation
   • Forecast stability metrics

Segment 2: Capacity Decision Framework

1. Capacity Metadata Integration

   • Flight capacity constraints
   • Cost structure analysis
   • Risk factor evaluation

2. 4-Factor Decision Lens Application

   • Demand stability assessment
   • Cost exposure analysis
   • Delay risk evaluation
   • Flexibility scoring

3. Strategy Generation

   • Route-specific capacity decisions
   • Utilization strategy assignment
   • Planning commentary generation

4. Results Validation

   • Decision logic verification
   • Business impact assessment
   • Framework scalability analysis

🚀 Future Enhancements

Planned Improvements

• Real-time capacity adjustment algorithms
• Advanced ML integration for pattern recognition
• Multi-modal transportation optimization
• Dynamic pricing integration with capacity decisions
• Enhanced visualization dashboard for planners

Research Directions

• Sustainability metrics incorporation in decision framework
• Customer satisfaction modeling for service level optimization
• Network-wide optimization beyond individual routes
• Predictive maintenance integration with capacity planning

📚 References & Resources

• FedEx Operations Research
• Supply Chain Optimization Best Practices
• Logistics Analytics and Capacity Planning
• Explainable AI in Operations Research

Contributors

Team Name: OneAboveAll Team Members: Partha Sarathy G Nithin bhargav G N Piritha

Opsium – FedEx Tricolor Challenge Finalist

For questions or collaborations, please reach out via GitHub Issues.