Ethan Le Sage

Predicting Talent Show Winners with Machine Learning

View the entire report here (1,807 KB).

For the final dissertation in my Computer Science BSc, my aim was to build a machine learning model capable of predicting the winners of I'm a Celebrity. It would do this using sentiment analysis on a large bank of tweets, and then by performing regression techniques on the results of sentiment analysis.

As part of this project, I developed a couple of novel techniques applicable to this specific problem set. One involved the creation of additonal training data with theoretical three-way votes that never actually occurred. The other was the usage of output post-normalisation, which enabled the use of three separate predictions to be used in tandem to generate one prediction for a three-way-vote.

The ideas explored in the months of model development are far too complex to express here (in fact they were difficult to express in a 50-page report!), so if you want to learn more please download the whole report with the link at the top of this page.