Data Scientist & Social Scientist · PhD in Political Science · Siegen, NRW, Germany
I spent years as a researcher and lecturer asking how institutions work, how power is distributed, and how to draw valid inferences from messy, real-world social data, mostly through research on local governance and decentralisation in Ghana. Data science is how I now answer those same questions at scale. I bridge the gap between raw quantitative datasets and actionable, human-centric strategies, from survey design to deployed model.
🟢 Open to Data Analyst / Data Scientist / Research Analyst roles (NRW, remote in Germany, or international). Available immediately.
- 🔭 Research: Research methods, local governance, and development cooperation, now combined with machine learning.
- 🛠️ Building: End-to-end data products: ETL pipelines, relational backends, predictive models, and RAG applications.
- 👯 Collaboration: Looking to partner with data scientists on social-impact projects.
- 💬 Ask me about: Research methods, survey data, data analysis, and development studies.
| Area | Tools |
|---|---|
| Languages & Core | Python (pandas, NumPy, scikit-learn), SQL (MySQL), SQLAlchemy |
| Machine Learning & AI | Supervised & unsupervised learning, regression, ensemble models, K-Means clustering, Retrieval-Augmented Generation (RAG), Ollama |
| Backend & Apps | Flask, Streamlit, Gradio |
| Cloud & Deployment | Google Cloud Platform (Cloud Functions, Cloud Scheduler, Cloud SQL), Hugging Face Spaces, Docker, Prefect, Git/GitHub |
| Data Visualization | Tableau, Matplotlib, Seaborn |
| Research Methods | Survey design, A/B testing & hypothesis testing, mixed methods, M&E, KoboToolbox, MAXQDA |
Domain: Machine Learning & Education Analytics
An end-to-end project that predicts which students are at risk of dropping out so advisers can intervene early.
- Classification: Random Forest compared against Logistic Regression and Decision Tree baselines on a stratified 80/20 split, handling a 62/38 class imbalance. 80% recall, 85% precision.
- Beyond prediction: Cohort clustering plus a local RAG layer (Ollama) to support advising, delivered as an interactive app (the-edumatch-app).
- View Project Repository ➡️
Domain: Computational Social Science & Survey Research
Tests the "dual allegiance" model from my own research on Ghanaian decentralisation, using Afrobarometer Round 9 survey data (n = 2,369).
- Inference: Paired t-tests, regional trust gaps and a logistic regression with odds ratios. Trust in traditional leaders (0.95) and elected councils (0.92) is statistically indistinguishable, and 60.6% disapprove of councillor performance.
- Machine learning: Random Forest robustness check with 5-fold cross-validation (ROC-AUC ~0.58), showing that demographics barely predict who leans traditional.
- App: Interactive Streamlit dashboard, prediction tool and data explorer, live here.
- View Project Repository ➡️
Domain: Data Engineering & Automated Cloud Architecture
A production-grade, end-to-end serverless ETL pipeline that automates data ingestion, transformation, and storage.
- Cloud Infrastructure: Fully serverless workflow using GCP Cloud Functions and Cloud Scheduler for automated execution.
- Database & Transformation: Relational schemas in Cloud SQL (MySQL), with Python scripts handling cleaning, structure mapping, and loading from REST APIs.
- Reporting: Pipeline output feeds Tableau dashboards of daily and seasonal trends.
- View Project Repository ➡️
Domain: Data Engineering & Database Design
A star-schema warehouse built from ~100k real Olist e-commerce orders spread across 8 relational CSV files.
- ETL & Modelling: pandas extract/transform/load into dimension and fact tables defined with SQLAlchemy ORM; idempotent full-refresh loads.
- Analytical SQL: RFM segmentation, delivery performance, revenue trends and category rankings using joins, CTEs and window functions.
- Quality: pytest suite for schema integrity and idempotency, plus a Streamlit dashboard on top of the warehouse.
- View Project Repository ➡️
Domain: Data Engineering & Pipeline Orchestration
An incremental ETL pipeline that normalises a 53-column flat file of ~180k supply-chain order lines into a relational schema.
- Incremental loading: Watermark-driven monthly batches that are resumable and safe to re-run.
- Orchestration: Prefect flow for retries and logging; SQLite by default, with an optional Postgres setup via Docker.
- Analytics & tests: SQL queries on delivery risk, profitability and customer frequency, backed by a pytest suite.
- View Project Repository ➡️
Domain: Database Design & Backend Development
A full-stack relational database and web backend that manages student records, staff operations, and administrative workflows for Good Foundation International School in Nsawam, Ghana.
- Stack: Flask, MySQL, SQLAlchemy.
- View Project Repository ➡️
Domain: Backend Development & Process Automation
A web application backend that automates member registration, database storage, and confirmation emails for Ghana Council e.V.
- Stack: Flask, MySQL, SQLAlchemy, automated email.
- View Project Repository ➡️
Domain: Data Analytics & Consumer Behavior
Evaluated the commercial impact of discount strategies on revenue, profitability, and seasonality for e-commerce retailer Eniac.
- Analysis: 15 months of order data analysed to weigh sales volume against profitability across customer segments and seasons.
- Recommendation: A targeted 5% discount model and an A/B testing strategy to protect margins.
- Data Storytelling: Publication-ready visualizations with Matplotlib and Seaborn.
- View Project Repository ➡️
Domain: Quantitative Research & Product Optimization
Analyzed an A/B/n experiment to optimize homepage click-through rate (CTR) for Eniac.
- Experimental Design: Checked the validity of user allocation across variants to ensure unbiased test conditions.
- Hypothesis Testing: Chi-square tests and confidence intervals to determine conversion-rate differences.
- Strategic Recommendations: Translated statistical outputs into clear design decisions.
- View Project Repository ➡️
Domain: Unsupervised Learning
K-Means clustering on Spotify audio features to segment a 5,000-song catalog into acoustically cohesive, listener-sized playlists.
Domain: Supervised Learning & Feature Engineering
A classification pipeline with Random Forest models to identify premium residential properties in the Ames Housing dataset.
Domain: Generative AI & Retrieval-Augmented Generation (RAG)
AI-powered RAG systems to retrieve, query, and analyze George Orwell's Animal Farm.
- Gradio + Hugging Face Spaces: LlamaIndex vector indexes with an LLM, deployed live for real-time semantic queries. Repository ➡️
- Local RAG with Ollama: Fully local Q&A and semantic document search. Repository ➡️
Domain: Databases
A concise reference guide to the foundational SQL commands every developer should know.
My academic work covers decentralisation, traditional authority, citizen participation, and ICT in public service delivery in Ghana, including coordination of three externally funded research projects (BMBF; BMZ via Engagement Global) and analysis of survey data from 550+ respondents that informed 5 policy recommendations.
- Anderson, P. K. L. (2024). Enhancing participation at the local level toward inclusive communities. In Inclusive Localities (Barbara Budrich). doi:10.3224/84743017
- Anderson, P. K. L. et al. (2022). FACIL-ICT: Facilitating the Interaction Between Citizens and Local Governments Through ICT. ZPE, University of Siegen. doi:10.25819/ubsi/10231
- Medium: Can We Predict Which Students Will Drop Out? and From Classrooms to Containers (@pakandersonn)
Full publication list on ORCID.
- Global Perspective: I am passionate about traveling, engaging with diverse communities, and exchanging cross-cultural ideas.
- Creative Outlets: Outside of data work, I find inspiration and analytical balance in cinema and music.
😄 Pronouns: He/Him