Evidence record · UDINUS · SDGs
- Unit
- Rektorat UDINUS
- Period
- 2024
- Goals
- Record
- 34da024d-757e-49ac-9c21-a0cd58972f26
- Published
- 2026-10-30
Universitas Dian Nuswantoro conducted a research activity on the use of the LUTanh activation function to optimize Bidirectional Long Short-Term Memory (Bi-LSTM) for earthquake forecasting. The activity involved researchers from UDINUS, namely Shidik, Guruh Fajar; Pramunendar, Ricardus Anggi; Kusuma, Edi Jaya; Saraswati, Galuh Wilujeng; Winarsih, Nurul Anisa Sri; Rohman, Muhammad Syaifur; Saputra, Filmada Ocky; Naufal, Muhammad; and Andono, Pulung Nurtantio. The activity addressed earthquakes as unpredictable and potentially catastrophic events whose consequences affect communities. It focused on the need for early earthquake prediction systems capable of forecasting earthquake location, magnitude, and depth.
The activity introduced LUTanh, described as a modified Rectified Linear Unit (ReLU) that combines the benefits of ReLU and Tanh activation functions. UDINUS applied and compared the performance of LUTanh in Long Short-Term Memory (LSTM) and Bidirectional Long Short-Term Memory (Bi-LSTM) algorithms for predicting earthquake disaster. The proposed method was also tested using a multiple input multiple output (MIMO) architecture for predicting the occurrence of earthquake disaster. This scope placed the activation function, recurrent neural network models, optimizer, and prediction architecture within one comparative research activity.
The evaluation results showed that LUTanh applied in the Bi-LSTM model, particularly when optimized with the Adam optimizer, consistently outperformed the LSTM counterpart. Error assessments of LUTanh in Bi-LSTM also consistently demonstrated lower average error scores than the original ReLU activation function, up to 4% for mean absolute error (MAE) and 3% for mean square error (MSE). Through this activity, UDINUS contributed research addressing earthquake forecasting and the development of methods relevant to the goal of sustainable cities and communities. The activity is therefore framed as institutional research supporting attention to earthquake-related risks through the stated forecasting approach.