Sistem deteksi dini untuk manajemen krisis penyediaan pupuk bersubsidi bagi petani padi (Studi Kasus di Kabupaten Banyumas, Jawa Tengah)

Abstract

Fertilizer supply crisis in Indonesia occur almost every year. This crisis can be prevented by early detection before the crisis situation occurred. Preventive action would be much better than remedial action. In this research, an Artificial Neural Network (ANN) based decision support for supply crisis of subsidized fertilizer distribution was developed. The Early Warning detection was developed with backpropagation (BP) learning’s methods. The constructing of the data’s input for ANN based on the fundamental parameters of supply crisis of subsidized fertilizer by using Exponential Comparation Method (ECM) and expert’s judgments with Analytical Hierarchy Process (AHP), consisting of the critical factor causing supply crisis of subsidized fertilizer, so this system be able to assess of intensity of the crisis with more fast, effective and efficient.Based on the “trial and error” test of ANN’s training process, the best network performance for BP learning’s method was obtained. The best network performance for BP was showed by the Mean Square Error (MSE) score of 0.00000000531 (5.31.10-9) at the 38th epoch, when the system used sigmoid bipolar for hidden layer and linear’s activation function for output neuron, Levenberg-Marquadt’s algorithm training, the momentum score was 0.05, the learning rate score was 0.05, and the minimum error was 0.0000005 with the network architecture of [9 60 30 1], that is, 9 neurons in an input layer, 60 neurons in a first hidden layer, 30 neurons in a second hidden layer and 1 neuron in an output layer. The BP was trained with 66 actual data and tested with 15 actual data, with 87% accuracy rate. From the test shows that the network has been able to recognize patterns of crisis. To get the improved accuracy of detection it is recommended for additional training with the latest data.