Back to Projects
O
Data Science / Analysis
Oil Production Forecasting
ML Engineer (Research)
TensorFlowXGBoostFastAPISHAPPython
About the Project
A PhD research project building an IoT-ML-Cloud framework for predicting NET oil production (bbls/d) across 4 Nigerian oil stations. The system combines IoT sensor data ingestion, machine learning model training with time-series features, and a cloud deployment pipeline. Models include regression for production prediction, classification for flow regime detection, and RNN models (LSTM/GRU) for sequential forecasting.
Key Highlights
- Built time-series forecasting pipeline with lagged features for oil production prediction
- Implemented multiple model types: regression, classification, and deep learning (LSTM/GRU)
- Developed FastAPI backend with Celery task queue for async model training
- SHAP explainability for understanding feature importance in production predictions
- Full-stack deployment with Next.js frontend for visualization
Technical Challenges
Oil production data is inherently noisy with sensor drift and missing values from IoT devices. Building reliable time-series features (lagged values, rolling statistics) while handling data quality issues across 4 different flow stations required robust preprocessing and validation pipelines.