Unsupervised Learning of Time-Series Classification Using Machine Learning Through Fertigation System

The implementation of smart fertigation systems in the agriculture industry is highly encouraged by the Malaysian government. Traditional manual monitoring and optimization approaches have proven unsatisfactory since they are time-consuming, unreliable, and impractical for huge farm regions. This study proposes the use of unsupervised machine learning clustering based on an automated monitoring device to predict the fertigation system requirements for optimizing eggplant growth, using ground-sensing parameters such as soil humidity, temperature, and electric conductivity. The goal is to improve eggplant development and production using the best-performing machine learning model. The study involves the collection of data from three units of sensors, which are stored in a database and then exported for feature extraction. The data is analyzed using k-means clustering and event identification processes to determine the best action for each cluster. By using ANOVA scoring method, it can determine the top 80 features in ranking out of 192 features. The study presents the implementation and examination of the information processing results of a week's worth of data. The system is operational and machine learning has been efficiently deployed. The study shows that random forest achieved 100% classification accuracy, support vector machine (SVM) reached 99.7%, while Naïve Bayes achieved 97.0%. In conclusion, this study contributes to the ongoing research on smart fertigation systems and their applications in agriculture, with a focus on unsupervised machine learning for time-series classification.