Master Thesis Project

Sport Activity Classification Using Classical Machine Learning and Time Series Methods

Python environment dependencies

Python version 3.8.16

channels:

• defaults

dependencies:

• ipykernel
• pandas
• seaborn
• scikit-learn
• threadpoolctl
• python=3.8
• numpy
• jinja2
• sktime
• tqdm

Testing environment

Device	Lenovo Yoga 920-13IKB
Processor	Intel(R) Core(TM) i5-8250U CPU @ 1.60GHz (1.80 GHz)
Installed RAM	8,00 GB
System type	64-bit operating system, x64-based processor

Development environment	Visual Studio Code, Jupyter extension
Version	1.74.3 (user setup)
Electron	19.1.8
Chromium	102.0.5005.167
Node.js	16.14.2
V8	10.2.154.15-electron.0
OS	Windows_NT x64 10.0.22621
Sandboxed	No

Jupyter Notebooks with preview are available in Kaggle

• Part 1: https://www.kaggle.com/code/jarnomatarmaa/masterprojectcode-cml
• Part 2: https://www.kaggle.com/code/jarnomatarmaa/masterprojectcode-tsc
• Part 3: https://www.kaggle.com/code/jarnomatarmaa/masterprojectcode-sac-analysis

Unfortunately, GitHub does not support Jupyter Notebook preview

Dataset recording devices

Data collection method (feature extraction from sensors)

Study pipeline

Adopted models

Sklearn

• k-Nearest Neighbors (kNN)
• Gaussian Naive Bayes (G-NB)
• Quadratic Discriminant Analysis (QDA)
• Logistic Regression (LR)
• Support Vector Machine (SVM)
• Decision Tree (DT)
• Random Forest (RF)
• Multilayer Perceptron (MLP)
• Linear Discriminant Analysis (LDA)

Sktime

• Time Series Forest Classifier (TSF)
• Supervised Time Series Forest (STSF)
• Random Interval Spectral Ensemble (RISE)
• Random Interval Classifier (RIC)
• Shapelet Transform Classifier (STC)
• KNeighbors Time Series Classifier (kNN-TS)
• Composable Time Series Forest Classifier (CTSF)
• WEASEL
• HIVE-COTEv1.0
• MUSE
• Column Ensemble (STSF-STSF-RIC)

Main results of the study

Standard Classical Machine Learning (S-CML)

Univariate and Multivariate Time Series Classification (U-TSC and M-TSC)

Result table (Compiled S-CML, U-TSC and M-TSC)