Selection of Machine Learning Algorithms for Limited Computing Device in Netted Melon Ripeness Detection

Abstract

The desired characteristics of melon fruit by consumers include sweetness, medium to large fruit size, thick flesh with appealing color and crisp texture, and relatively long shelf life. Predicting the harvest time of the fruit is essential in achieving the best harvest results. evolving technologies such as agricultural robots can be utilized as effective solutions and means. The utilization of automated robots in melon harvesting serves as a relevant example of implementation. These agricultural robots require a system capable of predicting the ripeness of melons for harvest. The study focuses on analyzing the performance comparison between two machine learning algorithms, Support Vector Machine (SVM) and Random Forest (RF), with the objective of determining the optimal choice when implementing them on limited computing devices. This study shows SVM and RF both have high accuracy values, 82% and 73%, respectively. Both of them also have fast computation times, with average inference times of 2.14 seconds and 2.15 seconds, respectively. The average CPU usage in the SVM algorithm is higher compared to the RF algorithm, at 17.80% and 15.48%. Eventhough SVM has higher accuracy rate, better precision, recall, and f score, but after conducting an independent 2-samples t-test on the inference time and CPU usage, it was found that there is no significant difference between SVM and RF. Both has good performance and good classification. The RF algorithm is recommended because it has a good accuracy rate, fast computation time, and less CPU resource usage.