Agus Wahyu Widodo, Titis Handayani, Fatma Agus Setyaningsih, Imam Nurhadi Purwanto, Ratno Bagus Edy Wibowo, Samingun Handoyo
In building a machine learning model, the quality of the model input has a significant role in producing the model with satisfactory performance. This study deploys the data mining feature selection to acquire the independent predictor as the input of multilayer perceptron neural networks (MLP NN) with tuning hyperparameters: node number in the hidden layer and the L2 penalty regularization. The complete predictor dataset is used to build the optimal MLP NN benchmark. Both MLP NN models have the same L2 penalty regularization of 0.05, whereas the node number in the hidden layer of 12 and 7 respectively for the dataset with the complete predictor and independent predictor. The performance evaluation of both MLP NN models in the testing data employing three metrics: accuracy, Mathew's Correlation Coefficient (MCC), and Area under curve (AUC) shows that the optimal MLP NN with independent predictors is not only producing the simpler model but also performing a slightly better than the optimal MLP NN with complete predictors. © Little Lion Scientific.
Informatics Engineering Department, Brawijaya University, Malang, 65145, Indonesia; Information System Study Program, Semarang University, Semarang, 50197, Indonesia; Mathematics Department, Yogyakarta State University, Yogyakarta, 94043, Indonesia; Mathematics Department, Brawijaya University, Malang, 65145, Indonesia; Data Science Study Program, Brawijaya University, Malang, 65145, Indonesia; Electrical Engineering and Computer Science-IGP Department, National Yang Ming Chiao Tung University, Hsinchu, 30010, Taiwan
