A Novel Approach for Classification in Resource Constrained Environments ARUN KUMAR ∗ , Indian Institute of Technology Indore, India ZHIJIE WANG, Invision AI, Canada ABHISHEK SRIVASTAVA, Indian Institute of Technology Indore, India Internet of Things’ deployments are increasingly dependent upon learning algorithms to analyse collected data, draw conclusions, and take decisions. The norm is to deploy such learning algorithms on the cloud and have IoT nodes interact with the cloud. While this is efective, it is rather wasteful in terms of energy expended and temporal latency. In this paper, the endeavour is to develop a technique that facilitates classiication, an important learning algorithm, within the extremely resource constrained environments of IoT nodes. The approach comprises selecting a small number of representative data points, called prototypes, from a large dataset and deploying these prototypes over IoT nodes. The prototypes are selected in a manner that they appropriately represent the complete dataset and are able to correctly classify new, incoming data. The novelty lies in the manner of prototype selection for a cluster that not only considers the location of datapoints of its own cluster but also that of datapoints in neighbouring clusters. The eicacy of the approach is validated using standard datasets and compared with state-of-the-art classiication techniques used in constrained environments. A real world deployment of the technique is done over an Arduino Uno based IoT node and shown to be efective. CCS Concepts: · Computer systems organization → Embedded systems; Redundancy; Robotics; · Networks → Network reliability. Additional Key Words and Phrases: IoT, Constrained Environment, Clustering, Classiication 1 INTRODUCTION There is widespread use of Internet of Things’ (IoT) deployments across domains nowadays. Such networks have the ability to assess a scenario and provide solutions in real time in smart home settings, agriculture, industries, and healthcare to name just a few areas. IoT data, therefore is heterogeneous and expensive to handle [1]. Authors in [2](EdgeMiningSim) explore approaches to handle such heterogeneous data in constrained and dynamic IoT environments. EdgeMiningSim is one such approach based on software engineering principles and is broadly a simulation-based data mining methodology. IoT deployments are nowadays also increasingly utilizing Machine Learning techniques to provide smart solutions. To do this, these networks capture data in real-time, send these to back-end clouds that are equipped with powerful learning algorithms, and then execute decisions received from the cloud. While this has been working well and has become the norm in most applications, the process of sending data back and forth seriously compromises the limited resources of IoT nodes and involves signiicant latency. This becomes a major bottleneck in applications that require instantaneous results and/or comprise ∗ Authors’ addresses: Arun Kumar, phd1701101005@iiti.ac.in, Indian Institute of Technology Indore, Simrol, Indore, Madhya Pradesh, India, 453552; Zhijie Wang, Invision AI, , Toronto, Canada, zhijie@ualberta.ca; Abhishek Srivastava, Indian Institute of Technology Indore, Simrol, Indore, Madhya Pradesh, India. Permission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for proit or commercial advantage and that copies bear this notice and the full citation on the irst page. Copyrights for components of this work owned by others than ACM must be honored. Abstracting with credit is permitted. To copy otherwise, or republish, to post on servers or to redistribute to lists, requires prior speciic permission and/or a fee. Request permissions from permissions@acm.org. © 2022 Association for Computing Machinery. 2577-6207/2022/7-ART $15.00 https://doi.org/10.1145/3549552 ACM Trans. Internet Things