In high-frequency airport operations, Turnaround Time (TAT) is a critical KPI. At major hubs like Dubai International (DXB), the "Below the Wing" operations (loading, fueling, catering) must be perfectly synchronized.
This project addresses the Manpower Forecasting and Scenario Modelling requirements for the Data Analytics Officer role. I developed a predictive system that identifies potential ground delays before they happen by analyzing the gap between required and deployed ramp agents during peak hub "waves."
- Database: MySQL (Relational storage for schedules and operational actuals).
- Language: Python (Pandas for ETL, Scikit-Learn for Predictive Modeling).
- ML Model: Random Forest Regressor (Selected for its ability to model non-linear operational bottlenecks).
- Security:
python-dotenv(Enterprise-grade environment variable management). - Dashboard: Streamlit (Interactive "What-If" simulation tool for Duty Managers).
- Data Engineering (SQL): * Designed a relational schema in MySQL to house flight schedules and operational performance.
- Created
v_manpower_gap_analysisview to calculate real-time staffing variances based on aircraft complexity (A380 vs. B777 vs. A320).
- Created
- Feature Engineering (Python): * Extracted Hub Peak Waves: Categorized arrivals into DXB's midnight (22:00-02:00) and morning (07:00-09:00) waves.
- Engineered the Staffing Gap feature to quantify the impact of under-allocation on departure punctuality.
- Predictive Modeling: * Trained a Random Forest model to predict turnaround delays in minutes.
- Result: Achieved a Mean Absolute Error (MAE) of 13.68 minutes, providing a baseline for identifying high-risk flights.
- Operational Dashboard: * Built a Streamlit application allowing users to adjust staffing levels and see the predicted impact on delays instantly.
- Non-Linearity: The model revealed that for wide-body aircraft (A380), the relationship between staff and delay is non-linear; missing the "minimum threshold" of 20 agents causes exponential delay growth.
- Peak Sensitivity: Flights arriving during "Peak Waves" are 3x more sensitive to staffing shortages due to shared ground equipment across gates.
- Incorporate Real-time Feeds: Integrate FlightRadar24 API for live arrival tracking.
- GSE Integration: Factor in Ground Service Equipment (GSE) availability into the prediction model.
- Refinement: Introduce "Deep Learning" (LSTM) to account for time-series dependencies in gate availability.
Developed by: Aklilu Abera