A Novel Determination of An Appropriate Clustering Quantity of A Water-Soluble NPK Nutrient Measuring System Based on K-means and SOM Methods

Recently, agricultural innovation has grown exponentially. Several technologies are used in smart farming to help increase productivity. Therefore, the development in electronic technology and the Internet of Things (IoT) devices to store data is vital. Determination of nutrients required for plant growth, such as nitrogen (N), phosphorus (P), and potassium (K) in soil or water, is key to the vertical or closed crop cultivation. However, the devices can measure the quantity of NPK nutrients with limited capacity and non-real-time system. This paper proposes a novel determination of an appropriate clustering quantity using K-means method and a prediction approach using the Self-Organizing Map (SOM) method to enhance the device capacity and real-time analytics. The number of 11,486 testing samples were collected from the water-soluble NPK nutrient monitoring system. Then, K-Means algorithm and unsupervised learning method was conducted to evaluate clustering of the data to train machine learning system. The data were analyzed and divided into two groups, resulting in a reduction in measurement errors and an improvement in reliability of test equipment. Consequently, Self-Organizing Map (SOM) was selected to analyze the same sample data to separate the weights from the values obtained from the K-Means process. The accuracy of SOM analysis was 0.5, enabling this approach to predict data during temperature changes. The best working value was predicted at temperature value of 26.47 °C, the red-light value at 5,048 Hz, the green light value at 9,006 Hz, and the blue light value at 2,246 Hz. Hence, machine learning can help improve measurement systems and reduce the failure of the cultivation system in the future.