Detection of the Possible Occurrence of Depressive Episodes in Patients with Multiple Sclerosis from Motor Activity Data

Project to obtain the Master's Degree in Data Science

Data preparation

Dataset

Depresjon dataset proposed by Garcia Ceja et. al. (2018), available at Simula Datasets

Install dependencies

$ pip install -r python-code/requirements.txt

Generating datasets for classification

$ scripts/generate_classification_datasets.sh

Spliting datasets

$ scripts/split_classification_datasets.sh

Feature extraction

$ scripts/feature_selection_all.sh -m rlo -v loo

or pass several models separated by comma

$ scripts/feature_selection_all.sh  -m nearcent,rlo,svm,sgd,rf,adaboost -v loo 

param -v can be either loo for Leave One Out validation or an integer specifying the number of folds in order to use k-fold cross validation

Note: feature generation is a very slow process and can be slower when using Leave one out (loo) validation

Training and testing models

$ scripts/build_classification_models.sh

Forecasting activity

$ scripts/build_forecasting_models.sh

Results analysis

Classification models results

$ jupyter notebook python-code/notebooks/classification_models_results_analysis.ipynb

forecasting models

$ jupyter notebook python-code/notebooks/forecasting_models.ipynb

Forecasting Classification results

$ jupyter notebook python-code/notebooks/deployment_results_analysis_mape.ipynb
$ jupyter notebook python-code/notebooks/deployment_results_analysis_rmse.ipynb

To see results in plots

$ jupyter notebook python-code/notebooks/deployment_results_analysis_plots.ipynb

References

• Garcia-Ceja, E., Riegler, M., Jakobsen, P., Tørresen, J., Nordgreen, T., Oedegaard, K. J., & Fasmer, O. B. (2018). Depresjon: A motor activity database of depression episodes in unipolar and bipolar patients. Proceedings of the 9th ACM Multimedia Systems Conference, MMSys 2018, 472–477. https://doi.org/10.1145/3204949.3208125