Dataset supporting the University of Southampton MPhil Thesis "Efficient Teacher-Student Architectures for Human Activity Recognition via Soft Labels and Binarization"

The data analyses three public datasets for Human Activity Recognition (HAR), with the original data directly downloadable from the Internet:
Daphnet Gait Dataset (Freezing of Gait): https://archive.ics.uci.edu/dataset/245/daphnet+freezing+of+gait

Opportunity Dataset: https://archive.ics.uci.edu/dataset/226/opportunity+activity+recognition

PAMAP2 Dataset: https://archive.ics.uci.edu/dataset/231/pamap2+physical+activity+monitoring

However, the data available there cannot be used directly and requires a series of data segmentation and preprocessing. What I have released here are the aforementioned three public datasets after undergoing a series of preprocessing steps.

The datasets have been preprocessed with Python, including sliding window cropping, removal of NaN rows, removal of time(ms), normalization, etc. They have been divided into Test, Train, and Validation datasets using mainstream methods and finally saved with Numpy for the convenience of users for quick deployment.

Daphnet Gait Dataset(Frozen of Gait):
https://archive.ics.uci.edu/dataset/245/daphnet+freezing+of+gait

This dataset is a binary classification dataset consisting of recordings from 10 participants diagnosed with Parkinson’s disease (PD). Dataset activities correspond to recognizing whether or not gait freeze occurs based on wearable acceleration sensors. The dataset was recorded in a lab environment with the subjects were instructed to carry out activities with a high likelihood of inducing freezing of gait, which is a common motor complication in PD.

Opportunity Dataset:
https://archive.ics.uci.edu/dataset/226/opportunity+activity+recognition

This dataset contains recordings from various wearables and environment sensors from four participants who carry out common kitchen activities, such as Open/Close Door, Dishwasher, and Fridge, via Inertial Measurement Units (IMUs) at 30Hz. Each participant is recorded in five different runs.

PAMAP2 Dataset:
https://archive.ics.uci.edu/dataset/231/pamap2+physical+activity+monitoring
The physical activity monitoring dataset is similar to the opportunity dataset, consisting of nine participants performing 12 kinds of daily physical activities, such as cycling, walking, sitting. The sensors used in the inertial measurement units (IMUs) include accelerometers, gyroscopes, magnetometers, temperature, and heart rate.

The data is accessible via CC BY license.