An analysis of the impact of uncertainty on the internet of things: A smart home case study

The Internet of Things (IoT) is the enabling technology for a range of smart domains and application areas. IoT consists of sensors, actuators, gateways and cloud infrastructures which together support the processing of information to enable decision makers to take appropriate actions in a given scenario. IoT infrastructures may, however, be associated with unreliable sensors and network infrastructures (e.g., due to incompleteness, inconsistency, inaccuracy and inaccessibility) which will lead to uncertainty in the decision-making process and as a result take inappropriate actions which may pose a threat to the safety and security of its users. The work presented in this paper focuses on analysing and evaluating the impact of different types of data imperfections and unreliable sensor data on the process of classification within an IoT instance applied in the domain of smart homes (SH). Four classifiers, namely Decision Tree (DT), Random Forest (RF), K-Nearest Neighbour (kNN) and Naïve Bayes (NB) were chosen for evaluation purpose. Classification results based on an openly available SH dataset were analysed and compared to evaluate and illustrate the performance and robustness of classifiers. The results indicated that it is difficult to declare a single classifier with high performance and robustness. Nevertheless, the RF classifier achieved the highest accuracy of 95.51%, 83.27% and 85.08% for attribute noise, missing and failure cases, respectively while NB achieved a highest accuracy of 79.5% in the case of class noise.