Work

Projects

Machine learning models, analyses and dashboards. Each card shows the approach, the evaluation metric and why the problem matters. Projects with a case study have a full write-up.

Data Science & Machine Learning0.95 R²

Ghana Maize Yield Prediction

An end-to-end pipeline predicting district-level maize yields in Ghana. It combines yield records with growing-season weather from the NASA POWER API, soil types and policy/pest indicators, then trains a Random Forest + Gradient Boosting ensemble.

Approach & impact

Approach

Standardised district names to merge yield, weather and soil sources; aggregated daily NASA weather into April–August growing-season features; added a previous-year yield lag, pest-outbreak and Planting for Food and Jobs policy indicators; trained a voting ensemble inside a scikit-learn pipeline.

Why it matters

District-level yield forecasts can support planning by farmers, agencies and food-security programmes in Ghana.

Read case study →Code on GitHub

Data Science & Machine Learning0.87 R²

Used Car Price Prediction

Regression models predicting car prices for ~11,900 vehicles from engine specs, fuel type, transmission, body style and year. Compares OLS, Lasso and Ridge, with Lasso used for feature selection.

Approach & impact

Approach

Started from a numeric-only baseline (test R² 0.57), then added one-hot encoded categoricals with rare-model grouping, raising test R² to 0.82. A cross-validated Lasso kept 86 of 182 features; Ridge reached R² 0.87 under 5-fold cross-validation.

Why it matters

A full regression workflow on messy tabular data, covering baselining, feature engineering, regularisation and honest validation.

Code on GitHub

Data Science & Machine LearningIn progress0.59 Churn recall

Customer Churn Prediction

Classifying which telecom customers will churn (7,000+ customers, imbalanced classes). Compares Decision Tree, Random Forest and XGBoost with SMOTE oversampling; the current Random Forest baseline reaches 78% test accuracy.

Approach & impact

Approach

Encoded categorical features, rebalanced the training set with SMOTE and compared tree-based models with 5-fold cross-validation. Next come stratified CV, hyperparameter tuning, threshold tuning for churn recall, and serving the model behind a small API.

Why it matters

Recall on churners matters more than overall accuracy, because a missed churner is a lost customer. The work focuses on improving that number.

Code coming soon

Analysis286 Respondents

Ghana Tech Ecosystem Survey 2025

Turned raw survey responses from 286 Ghanaian tech professionals into a 4-page interactive Power BI dashboard covering roles, skills gaps, AI tool adoption, income and ecosystem needs.

Approach & impact

Approach

Cleaned and de-duplicated responses (279 unique respondents), reshaped 43 questions into a 7,425-row long format, and built drill-through dashboard pages with 18 calculated measures.

Why it matters

Findings such as fintech accounting for 41% of respondents, an 11.5% female share and better-paying local jobs as the top need (56%) give policymakers, investors and trainers a clear evidence base.

Code on GitHub

Dashboards & BI35% Junior–senior pay gap

Data Science Job Market Analysis

Cleaned a 2,000+ row job postings dataset and benchmarked salaries across 10+ job titles and 5 industries, surfacing the most in-demand skills and the pay gap between junior and senior roles.

Approach & impact

Approach

Handled duplicates and missing values with filters and pivot tables, benchmarked salaries with VLOOKUP, IF and AVERAGEIFS, and summarised the results in an interactive dashboard.

Why it matters

Identifies Excel, SQL and Python as the top three requested skills, which is practical guidance for job-seekers and hiring teams.

Code on GitHub

Dashboards & BI8,523 Items analysed

Blinkit Grocery Sales Analysis

SQL analysis and an interactive Power BI dashboard on 8,523 grocery items sold across 10 Blinkit outlets, covering 1.2M in total sales broken down by item type, fat content, outlet size and city tier.

Approach & impact

Approach

Standardised inconsistent category labels in SQL (five spellings of "Low Fat" and "Regular" merged into two), computed KPIs (total and average sales, item count, average rating), and broke sales down by item type, outlet tier and fat content with grouped and pivoted queries before building the Power BI dashboard.

Why it matters

Shows where revenue comes from (fruits & vegetables and snack foods lead), giving a retail team a clear view of which categories and outlet types to prioritise.

Code on GitHub

Analysis

Income & Demographics Analysis

Analysed income distribution across education, occupation, gender, race, work hours and country, revealing clear socioeconomic patterns in earning disparities.

Approach & impact

Approach

Analysed income distribution across demographic and work variables with Pandas, using Seaborn and Matplotlib visualisations to surface patterns in earnings.

Why it matters

Translates a large demographic dataset into clear, visual findings about who earns more, and why.

Code on GitHub