An efficient approach to kNN algorithm for IoT devices

K nearest neighbor is a popular method for classification, but it suffers from high runtime and space complexity. Various advancements have been made to improve classification accuracy and reduce computation time and memory. Algorithms such as kd tree kNN progress towards lower runtime complexities, but there is still scope for further improvements. This work presents an approach that leverages the spatial arrangement of the points visualized in spherical partitions to classify the data using the nearest neighbors. Our proposed method has a constant prediction time and a constant model size and produces results with accuracy similar to kd tree kNN, making it ideal for IoT devices. We compare and analyze the results of our approach with kd tree kNN on publicly available datasets of varying sizes.