Evaluating KNN Performance on WESAD Dataset

In this paper performance of KNN models are evaluated by changing K-fold cross validation parameter and total number of nearest neighbors while classifying WESAD dataset using sklearn library of python programming language, in order to finalize best possible number of nearest neighbors. Performance of KNN models drastically change when total number of nearest neighbors are modified irrespective of the dataset. Consequently for KNN based machine learning applications, tradeoff between optimum performance and computational cost is achieved by limiting total number of neighbors and hence controlling complexity of the model. Thus less computationally expensive KNN models can be directly implemented on raspberry pi, multicore microcontrollers, and low power IoT devices for classifying sensor data on portable embedded systems.