Hyperparameter Tuning using GridsearchCV on The Comparison of The Activation Function of The ELM Method to The Classification of Pneumonia in Toddlers

Pneumonia is a disease that is susceptible to attack toddlers. According to data from the Ministry of Health, the cause of under-five deaths due to pneumonia is number 2 of all under-five deaths. In Kalimantan, forest fires are one of the causes of the high number of pneumonia cases. Knowing the symptoms of the disease is very important, considering that sufferers often do not know that they have been exposed to Pneumonia because the symptoms that appear are just ordinary pain. In this study, the classification of Pneumonia and Non- Pneumonia Cough was carried out based on symptom factors. The dataset used in this study is the Poly MTBS at the Martapura Timur Health Center. The classification method used in this research is Extreme Learning Machine (ELM). The classification process starts from SMOTE upsampling, this is done to balance the classes, because the amount of data between classes used is not balanced. Then hyper tuning the parameters is done using GridsearchCV on the hidden layer neurons, to determine the best parameters that will be used as recommendations in the classification process. At the classification stage using the ELM method by comparing the activation functions of Binary Sigmoid, Sin, Hard Limit, Triangular Basis, Radial Base, Linear, and Bipolar Sigmoid by comparing test datasets 90:10, 80:20, 70:30, 60:40, and 50:50. This study provides the best performance results on the use of the Triangular Base activation function with 86.36% accuracy, 85% precision, 100% recall and 92% F1 Score, training data ratio, and testing 90:10 and 3 hidden layer neurons.