Through sensor networks, agriculture can be connected to the IoT, which allows us to create connections among agronomists, farmers, and crops regardless of their geographical differences. Faulty sensor detection is critical in IoT. When a sensor becomes faulty, missing data and/or bad data is provided to the control and management systems, which may lead to potential malfunction or even system failures. Because of this, a sensor fault detection mechanism must be implemented in an IoT system to eliminate this potential fault. This paper focuses on the implementation of a faulty sensor detection mechanism using data correlation among multivariate sensor readings, which is called Multivariate Faulty Sensor Detection Mechanism (Multi-FSDM) in a smart agriculture system. The smart agriculture system is attached with multi-variate sensors, which are moisture, temperature, and water sensor. These sensors are connected to Arduino UNO, which is equipped with an ESP8266 Wi-Fi module for internet connectivity. ThingsBoard is selected as the IoT cloud platform. The sensor readings are collected periodically and send to the cloud via the internet. Multi-FSDM calculates the correlation between each sensor reading to determine the health condition of each sensor. When all sensors are in good condition, all sensor readings are correlated with each other. However, when any sensor becomes faulty, sensor readings become uncorrelated. Once uncorrelated sensor readings occur, this means a faulty sensor is detected. Based on the findings, it is proven that Multi-FSDM can detect each sensor state on the smart agriculture system either in a good or faulty condition. When a sensor becomes faulty, Multi-FSDM detects and determines the faulty sensor successfully.

IoT devices are lightweight and have limited computational capabilities often exposed to harsh environments, which can cause failure on the IoT devices [1, 2].  The failure on the IoT devices is also caused due to limited battery life, hardware failure or human mistakes. Sensor faults can be categorized under one type of hardware failure, such as sensor burn, reduced sensor sensitivity and malfunctioned sensors.  Any faulty on the IoT devices can cause a problem on the overall operation of the IoT system. Traditional ways in the management of IoT devices is a maintenance officer require to check each device every day  [1, 3]. Any faulty devices found needs to be fixed or replaced. This traditional method is not practical and very challenging especially in the management of a large scale deployment of IoT consist of hundreds or thousands devices. Because of this, we proposed a faulty sensor detection and identification mechanism using multivariate sensors. Two methods of decision making are introduced in detecting faulty sensors, which are logical and correlation method that implemented in smart parking system and smart agriculture system accordingly. The logical method compares state of all sensors (ultrasound, IR and hall-effect) in the smart parking system either a parking lot is occupied or available, and then determine the condition of the sensors. The drawback of this method is not able to detect faulty sensor properly for a constant fault, which the sensor reading remains the same value. The correlation method calculates the correlation between all sensors (soil moisture, soil temperature and soil water) in the smart agriculture system. This method uses a moving window technique to calculate the correlation for all sensor over time. Any incomparable and uncorrelated sensor readings means a presence of faulty sensors. Based on the experiment results, the findings shows that the proposed faulty sensor detection mechanism is working properly in detecting faulty sensor in a timely manner.