This is the code written towards completing a research project titled: "A study of how outlier detectors can accurately authenticate multiple persons using the heart rate from consumer-grade wearables". This was completed for TU Delft with the suppervision of David Tax, Arman Naseri Jahfari, Ramin Ghorbani.
1vN_GMM.py and 1vN_SVM.py are the ones that generated:
- results/GMM/all_transformations/result_1vs11_1h.csv, results/GMM/all_transformations/result_1vs11_3h.csv
- results/OneClassSVM/all_transformations/result_1vs11_1h.csv, results/OneClassSVM/all_transformations/result_1vs11_3h.cs Which contain every result per each of the 100 combinations of 100 people. The folowing final results, that take the mean and standard deviation AUC are presented below and calculated using results_combining_and_graph_creation.ipynb:
- results/GMM/all_transformations/final_1vs11_1h.csv, results/GMM/all_transformations/final_1vs11_3h.csv
- results/OneClassSVM/all_transformations/final_1vs11_1h.csv, results/OneClassSVM/all_transformations/final_1vs11_3h.cs
These was generated by GMM_one_model.py and SVM_one_model.py and include the following files:
- results/GMM/all_transformations/result_NvsN_one_3h.json
- results/OneClassSVM/all_transformations/result_NvsN_one_3h.json
These was generated by GMM_multiple_model.py and SVM_multiple_model.py and include the following files:
- results/GMM/all_transformations/result_NvsN_multiple_3h.json
- results/OneClassSVM/all_transformations/result_NvsN_multiple_3h.json
The rest of the results under the folder results are experimental and important in deciding the final setup of the final experiments that were important in the paper. Many of these results contain values generated by faulty code. I do not advise on using them as they are.