Performance and Cost of Machine Learning Models for Seizure Detection in an Adaptive e-Health IoT platform over Edge and Fog nodes
Résumé
This work focused on the evaluation of some machine learn- ing (ML) models and their application in e-health within Internet of Things (IoT) platforms used for the detection of seizures or epileptic episodes. The evaluation was based on two groups of metrics; the first group consists of statistical validation and the second group sought to measure the cost and computational complexity of the models, the two groups of metrics were applied in the training and validation phases. The results obtained can determine relevant factors for the selection of ML models, either based on the statistical and intrinsic efficiency of ML models, or on their suitability to be implemented in IoT under the crite- ria of cost and computational complexity, which are directly related to their energy consumption. The evaluation scenario was defined under an architecture with Edge, Fog and Cloud (EFC) layers, where the models were implemented, initially in the cloud layer, then in the fog and edge layer. The results highlight that GBC and XGBC models present better performance when run from the cloud; LR, NB and SNN models can be trained from fog nodes and, finally, SLR and MLP can be implemented and used from edge nodes. MLP especially presents a good balance be- tween low computational cost and high accuracy in seizure detection.