George Kampolis

Hi, I am an Engineering Doctoral candidate at the University of Strathclyde, Faculty of Engineering in Glasgow, UK. In particular, I'm part of a cohort of, and affiliated with the Centre Doctoral Training for Wind and Marine Energy Systems & Structures (CDT-WAMESS). I'm working on using data science, machine learning and signal processing in the renewable energy sector, focusing on remaining useful life prediction and lifetime extension of wind turbine drivetrains.

Previously, I completed my MSc in Data Science at Robert Gordon University and before that, my MSc in Renewable Energy & Environmental Modelling at the University of Dundee. It all started with my undergraduate MEng in Mineral Resources Engineering (focusing on energy & petroleum engineering), completed at the Technical University of Crete.

I like to focus on getting the meaning and stories from numbers, be it data science or engineering interpretation (or both!), and optimizing systems and solutions. And of course I'm all about science jokes—sometimes I'm a nerd like that.

My favourite question is "Why?"


Below are a few examples of projects with code I’ve written — see my GitHub profile for more.:

  • WandererID, an open-source solution for automatic zooplankton classification using R, developed for Marine Scotland Science.

    A real-life data set was utilised with zooplankton measurements to create tuned classification solutions for multi-class problems using XGBoost, Random Forest and as a surrogate model, k-NN. The resulting model based on XGBoost outperformed previous solutions by 3.12% and above in terms of accuracy, while also creating a prototype system for operationalization of the classifier to achieve productivity gains (which is available in WandererPrototype). Finally, a notebook introducing the dataset itself with some basic exploration is accompanying the R scripts, which is available here. This is the code component of my MSc in Data Science dissertation.

  • TitanicAPI, a project in R exploring Model-as-a-Service via API creation.

    For demonstration purposes, created a Naive Bayes classifier on the Titanic data set using the mlr and plumber packages, which is then served as an online API via a hosted Docker container (tested on Microsoft Azure). The project can serve both as a straightforward exercise and as a springboard for future projects.

  • ChilWind,a project in R which explores statistical wind speed forecasting for power generation with the primary aim to see if better forecasts can be generated before feeding data into a Numerical Weather Prediction models for longer forecasting horizons.

    Cleaned a data set of wind speeds (6+ million measurements) obtained at Chilbolton, UK, with annual and diurnal seasonality. Constructed hourly time series and forecasted with various methods incl. ETS, (s)ARIMA and STL approaches. Showed clear improvement (in terms of mean absolute error) over the benchmark of simple persistence for a forecasting horizon of 48 hours.


Feel free to contact me if you require the most up-to-date version.