This project evaluates the performance of various classification algorithms on a dataset of gym exercises recorded using time series data from the accelerometer and gyroscope sensors of a smartphone device. The primary objective is to determine if you can accurately classify gym exercises based on the sensor data collected. The project compares a number of common machine learning algorithms such as Random Forest, k-Nearest Neighbours with Euclidean Distance and Naïve Bayes, as well as some time series specific algorithms, including Rocket and Time Series Forest, on a range of different variations of the dataset. These variations include different attributes such as using univariate data from a single axis or using multivariate data by combining data from multiple axes. Prior to analysis, the time series data is pre-processed to use same-length normalised time series data of 10 seconds. The results of this project show that when using multivariate time series data that uses both the accelerometer and gyroscope data, an accuracy of 99.2% can be achieved using the Rocket classifier.

If you are interested in reading my project report check out the pdf: here