Featured · final-year project
NGX Volatility Prediction System
A machine-learning system that predicts volatility for stocks on the Nigerian Exchange, with an interactive analytics dashboard. It forecasts how much stocks swing rather than trying to predict their price, and explains its predictions with SHAP.
PythonXGBoostRandom ForestSVRSHAPStreamlit
Read the case study
- Problem
- Investors and analysts need a view of risk, and volatility is a more forecastable quantity than price.
- Data
- Daily OHLCV history, 2020–2025, for fifteen NGX stocks across banking, telecoms and technology.
- Models
- Ridge Regression, Random Forest, XGBoost and Support Vector Regression, plus GARCH-related experiments.
- Evaluation
- RMSE and R² per stock. Results differ by stock, so no single score describes the whole system.
- Interpretability
- SHAP shows which features drive the tree-based models’ forecasts, so the output is not a black box.
- Application
- A Streamlit interface for exploring forecasts. I have not published a live link here.
Time series · riskNGX Forecasting & Anomaly Detection
Time-series forecasting with ARIMA and Prophet, plus anomaly detection on the NGX All-Share Index. It extends my undergraduate research on NGX volatility and adds SQL and a Power BI view for risk monitoring.
ARIMAProphetAnomaly detectionSQLPower BI
Machine learningFraud Detection Model
Detects fraudulent credit card transactions in a highly imbalanced dataset where only 0.17% of transactions are fraud. It uses SMOTE and Random Forest, with EDA and a comparison of models judged on Precision, Recall, F1 and ROC-AUC rather than accuracy.
PythonScikit-learnSMOTERandom Forest
Customer analyticsCustomer Segmentation & CLV
Groups customers with RFM analysis and K-Means clustering, then estimates customer lifetime value and churn risk. Results are delivered through SQL queries and a Power BI dashboard.
RFMK-MeansCLVChurn riskSQLPower BI
Recommender systemMovie Recommender System
Collaborative filtering with matrix factorization written from scratch, including cold-start handling. It is evaluated with MAP and NDCG and comes with a FastAPI and HTML demo.
Collaborative filteringMatrix factorizationMAP / NDCGFastAPI
Deep learningSpeech Recognition (DeepSpeech-Lite)
A speech-to-text recognizer built from scratch in PyTorch using a CNN, a bidirectional GRU and CTC loss. It is trained on LibriSpeech and includes a Gradio demo.
PyTorchCNNBiGRUCTCLibriSpeechGradio
Business analyticsNetflix Data Analysis
A business-focused exploratory analysis of Netflix’s content strategy covering growth trends, geographic opportunity, genre analysis and content aging. It uses SQL and Power BI and ends in actionable recommendations.
PythonSQLPower BIEDA