Data Science Portfolio#
Welcome! I’m Alun, a Statistical Methodologist at the UK Office for National Statistics with a PhD in Computational Physics. This portfolio showcases projects spanning machine learning, computer vision, time series analysis, NLP, small-area estimation, and applied statistics.
Featured Projects#
Power Laws and Popularity
What 50,000 synthetic Spotify tracks can and cannot tell you: leakage, grouped validation, and a model ladder that stops when capacity stops paying.
Finding Cancer in Ultrasound Images
A model ladder on public BreastMNIST ultrasound — from PCA-logistic and HOG forests to a small CNN and ResNet-18 transfer, with Grad-CAM explanations and a full metric dashboard.
Small Area Estimation for UK Local Authorities
Fay–Herriot EBLUP and a Bayesian hierarchical model for APS unemployment rates, validated against ONS published model-based estimates.
Literary NLP: Sentiment Across Classic Novels
Sentiment, emotion, term frequency, word clouds, and character presence mapped through eight public-domain novels — from Dracula to A Christmas Carol.
ODI Win Probability: From DLS to Data
Ball-by-ball win probability for one-day cricket — simplified DLS resource baselines, logistic regression and LightGBM models, and interactive chase charts.
Time Series Analysis: Classical Methods
A rigorous tour of classical forecasting — exponential smoothing, Fourier methods, ARIMA/SARIMA, and state-space models — applied to benchmark datasets from Air Passengers to Mauna Loa CO₂.
UK Urban Systems Network
A UK replication of the French systèmes urbains régionaux cartogram: TTWA nodes, census commuting flows, and Nystuen–Dacey dominant-flow partitions built in R.
UK Terrain & Rugby Geography
Raster and vector spatial analysis using the tidyterra R package. Combines UK elevation data with English Premiership rugby club locations to explore the geography of the game.
Titanic Survival Analysis
A comprehensive walkthrough of classification modeling—from messy data to deployed predictions.
French Rugby Voronoi Analysis
Geospatial analysis of French professional rugby teams using Voronoi diagrams. Showcases computational geometry, interactive mapping, and sports analytics.
Beatles Discography Visual Story
Interactive music analytics using The Beatles Spotify dataset, including writer-level popularity and sunburst views of the full discography.
Technical Stack#
Languages & Tools: Python • R • SQL • Statistical Modelling • Machine Learning • Computer Vision • NLP • Geospatial Analysis • Time Series • Seasonal Adjustment • Reproducible Analytical Pipelines
Key Libraries (Python): scikit-learn • PyTorch • pandas • numpy • matplotlib • plotly • nltk • geopandas • shapely • folium • pymc
Key Libraries (R): rjd3/JDemetra+ • KFAS • xts • tidyverse • Shiny • Quarto
About This Portfolio#
This site is built with Jupyter Book, which allows me to combine narrative explanations with executable code. Each project tells a complete story—from initial data exploration through modeling to final conclusions—while maintaining reproducibility and technical rigor.
All code is open-source and available on GitHub. Feel free to explore, learn, and adapt the analyses for your own work.
Last updated: 2026