Improved Hybrid Machine Learning User Behavioural Model for Secured Smart Homes

Abstract— This paper proffers a secured and cost effective
model for user behavioural analysis (UBA) in smart homes. The
model was built with support vector machine algorithm on a
generated dataset on constructed prototype of smart home. The
prototype had motion sensors and actuators for data collection and
switching on and off of things in the home. Another model called
human activity recognition (HAR) was developed with Gaussian
naïve bayes to interface with UBA for collection of information
from the home. The processor of the system is raspberry pi 4
computer board that runs on Linux-based Raspbian Operating
System. The two models earlier mentioned were integrated in the
processor. The information generated from the system is processed
in the processor and disseminated to things in the home. The Blynk
Mobile App was used to interface with the integrated system which
enables communication between the system and the end user.
Experiments were performed on the constructed prototyped smart
home, the results depict a convincing prediction of normal and
abnormal situation which contributed immensely to the security of
the home. On model evaluation, out of the twenty experiments that
were performed, the True Positive value was 11, False positive
value was 0, True negative value was 8 and False negative value
was 1. This resulted in prediction accuracy of 95%, precision of
100%, Recall of 92% and F1_Score of 96%. This obviously proved
that the developed system can provide security to about 95% in
smart home which is good and will be acceptable in all standard.
Keywordst; Internet of Things, Smart home, Human Activity
Recognition, Security, User Behaviour Analysis, Anomaly detection.