Penerapan Artificial Neural Network Untuk Memprediksi Error dalam Perancangan Aplikasi Monitoring Tetes Cairan Infus

Abstract

Decision-making in error prediction during the design process of infusion drip monitoring applications plays a crucial role. Designing this application is necessary because manual monitoring by medical staff is prone to errors and inaccuracies. Therefore, the need for accurate predictions in both planning and error management must be further investigated. This research discusses the benefits of the Artificial Neural Network (ANN) methodology in addressing error values in infusion drip monitoring applications during the design process. ANN is chosen for its ability to handle data complexity and non-linear patterns in infusion drip rates. Errors in infusion dosage can be fatal, ranging from patient instability to severe complications. Designing infusion drip monitoring applications automates the process and ensures accuracy, reducing the workload of medical staff and enhancing patient safety. This application also allows for more consistent and real-time monitoring, enabling quicker medical intervention when issues arise. The ANN methodology used includes both forwardpropagation and backpropagation, employing a binary sigmoid activation function with a learning rate of 0.03 and a maximum epoch setting of up to 1000. The research results indicate that the model-building procedure consists of several stages: (1) Determining input based on infusion drip rate readings. (2) Splitting the data into training and testing datasets. (3) Normalizing the data. (4) Building the forwardpropagation and backpropagation algorithm by determining the number of hidden layers, optimal input, and model weights. (5) Denormalizing the data. (6) Testing the model's accuracy. The ANN simulation revealed the best network structure using a 3-40-1 configuration (3 input variables, 40 hidden layers, and 1 output). The results achieved an average error prediction accuracy of 98.6%.