Optimisasi Berkelanjutan Menggunakan Machine Learning untuk Forecasting Performa Pertumbuhan Tanaman

Abstract

The distribution of data generated at each iteration in the continuous optimization process tends to result in premature convergence because optimum points are found at the beginning of the iteration so that true optimum conditions cannot be achieved. A method is needed that can find optimum points at each iteration in a continuous optimization process. The challenge of this research is how to determine the optimum points in each iteration of the range of variables generated. In this study, a multi-linear regression approach was used to forcasting the variables generated at each iteration, then the linear regression model was optimized using a neural network method approach. Implemented and observed on the growth morphology of chili plants with a sample of 100 observed during the 100 day growth period. With a percentage of 70% training data and 30% testing data, the research results obtained were that using the Relu activation function had a very ideal value compared to the Tanh, Sofplus, Elu and Sigmoid activation functions. When compared with the Time Series method with an MAE value of 4.62, this value is a very good value, while for the time series method it is still high at 8.6. Likewise, the RMSE and MAPE measurement values of 16.36 and 36.53 are very good