Selection of Soil Features for Detection of Ganoderma Using Rough Set Theory

Ganoderma boninense (G. boninense) is one of the critical palm oil diseases that have caused major loss in palm oil production, especially in Malaysia. Current detection methods are based on molecular and non-molecular approaches. Unfortunately, both are expensive and time consuming. Meanwhile, wireless sensor networks (WSNs) have been successfully used in precision agriculture and have a potential to be deployed in palm oil plantation. The success of using WSN to detect anomalous events in other domain reaffirms that WSN could be used to detect the presence of G. boninense, since WSN has some resource constraints such as energy and memory. This paper focuses on feature selection to ensure only significant and relevant data that will be collected and transmitted by the sensor nodes. Sixteen soil features have been collected from the palm oil plantation. This research used rough set technique to do feature selection. Few algorithms were compared in terms of their classification accuracy, and we found that genetic algorithm gave the best combination of feature subset to signify the presence of Ganoderma in soil.