Portfolio Classic Machine Learning

A collection of notebooks illustrating projects which engineer features for machine learning models

Reducing algorithmic bias in Audio Classification - Full machine learning discovery process.ipynb

Describes a full discovery process from problem identification, feature engineering, model selection and training. The problem: using short, noisey, audio clips recorded in 36 different urban locations

• 1. Classify voices contained in the audio as Male or Female.
• 2. During model selection and training, as much as possible, reduce the difference in accuracy between male and female voices.

Audio Classification - indoors or outdoors - Full Machine learning process.ipynb

Describes a full discovery process from problem, feature engineering, model selection and training. The problem: using short, noisey, audio clips recorded in 6 locations

• 1. Classify background noise as being recorded indoor or outdoors
• 2. During model selection and training, as much as possible, reduce the difference in False Postitives between classes

The projects goal was not to maximise accuracy but to demonstrate methodology in feature enginnering, model selection and model training. We were given no resources or special training in working with audio signals before hand, so the project encouraged independent exploration of feature engineering techniques and model training.