ABSTRACT Earthquake prediction remains one of the most challenging tasks in natural hazard research due to the complexity and heterogeneity of seismic processes in space and time. To address this, we propose a deep learning (DL) framework based on graph convolutional and recurrent neural networks (RNNs) to forecast the maximum seismic magnitude in different regions of Chile. Using a cleaned and spatially segmented catalog of seismic events, we construct a graph where each node represents a seismic cluster derived from K‐means clustering, with edges reflecting spatial proximity. Two models are evaluated: a standard Long Short‐Term Memory (LSTM) network and a hybrid Graph Convolutional Network‐LSTM (GCN‐LSTM), which incorporates both temporal dynamics and spatial dependencies. Our results show that the GCN‐LSTM model significantly outperforms the simple LSTM in terms of F1‐score and recall, especially in regions with complex seismic activity. This demonstrates the advantage of graph‐based neural models in capturing spatial correlations and improving earthquake magnitude prediction at a regional scale.

Data-driven earthquake magnitude prediction models typically rely on sliding-window representations of seismic catalogs, yet the choice of window length and stride is often made heuristically.This paper investigates how an enhanced sliding-window construction and its systematic optimisation affect next-event magnitude prediction across three tectonically distinct regions.We compile multi-decadal earthquake catalogs from the USGS ComCat for Java-Bali, Chile, and Iran (1970-2020) and construct supervised samples by segmenting each regional catalog into overlapping sequences of the last events.We then treat the window length and stride as explicit hyperparameters and select an enhanced configuration ( * , * ) on the Java-Bali validation set using a compact model.On top of this representation, we evaluate BiLSTM-only and CNN-BiLSTM architectures alongside naive (mean and persistence) and window-based deep learning baselines.In the Java-Bali region, the enhanced sliding-window BiLSTM achieves the lowest test errors (MAE 0.47, RMSE 0.62), outperforming fixed-window and short-history baselines and indicating that carefully tuned temporal representation can be as important as architectural complexity.Cross-regional experiments, in which a model trained on Java-Bali is transferred without fine-tuning to Chile and Iran, reveal higher errors and occasionally negative 2 values, reflecting the difficulty of next-event magnitude prediction under strong domain shift.Nevertheless, the enhanced representation typically performs competitively with, and often better than, naive and fixed-window baselines.Overall, our results highlight the critical role of sliding-window design in catalogbased magnitude prediction and provide a multi-regional benchmark for future work on domain-adapted and probabilistic deep learning models.

This study presents a hybrid predictive model that integrates Artificial Neural Networks (ANN) and Particle Swarm Optimization (PSO) to forecast earthquake timing and magnitude in Saman, Iran, with a strong focus on vibration signal analysis and dynamic measurement. The offered model implements 12 vibration-based input features, including peak ground acceleration (PGA), shear wave velocity, and spectral intensity, all of which are derived from seismotectonic and accelerometer data. PSO optimizes ANN weight initialization. This approach enhances the model’s ability to accurately represent seismic wave dynamics, making it well-suited for applications in vibration engineering. The dataset consisted of historical seismic records and was divided into 80% for training and 20% for testing. The ANN-PSO model outperformed conventional ANN and SVM, achieving denormalized RMSE of 0.152 (magnitude) and 0.189 (timing window likelihood error), MAE of 0.118 and 0.147, R² of 0.958 and 0.941, and Pearson r of 0.979 and 0.970 across 20 runs (MSE 0.023 on normalized scale, equivalent after denormalization). Thus, it was identified as a robust tool in the fields of vibration-based seismic forecasting, structural health monitoring, and mechanical reliability analysis in tectonically active regions.

Indonesia is a country with high seismic activity due to its location at the convergence of three major tectonic plates. This condition creates a strong need for earthquake pattern analysis and magnitude prediction to support disaster mitigation. This study aims to cluster earthquake data using the Density-Based Spatial Clustering of Applications with Noise (DBSCAN) algorithm and to predict earthquake magnitude using the Random Forest algorithm optimized through hyperparameter tuning. The Indonesian earthquake dataset was obtained from Kaggle with a total of 92,887 valid entries. The DBSCAN clustering results revealed several active seismic zones, particularly in Sumatra, Java, Sulawesi, and Papua. The comparison of R² between the Baseline Random Forest and the Tuned Random Forest shows a significant improvement after the parameter tuning process. The Tuned Random Forest model achieves an R² value of 0.478, which is higher than the Baseline Random Forest's 0.442. This indicates that the tuned model is better able to explain the variance in the data and provides more accurate predictions.

Hybrid LSTM model with efficient hyperparameter tuning for earthquake magnitude prediction in Turkey

ABSTRACT Multi-precursor analysis, along with multi-method analysis, has made it possible to detect a large number of LAI seismic anomalies in the study of strong earthquake-affected areas. It seems necessary to use fusion-based methods to combine the detected anomalies and earthquake parameters (magnitude, time and location) obtained from the analysis of different precursors and extract the final earthquake parameters along with their uncertainty. This study is carried out in two phases, which are: 1) the deviation values of about 1030 LAI anomalies detected using 20 implemented predictor algorithms around the time and location of 23 powerful earthquakes that occurred in recent years have been considered. Various functions were fitted on the collected data, including the anomaly intensity, and real magnitude of the earthquake, and a functional model was developed to estimate magnitude parameter with RMSE of about 0.53. 2) Using the Dempster-Shaffer (DS) Theory, the predicted earthquake magnitudes from different precursors for three strong earthquakes including Turkey (February 06, 2023), Myanmar (March 28, 2025) and Russia (July 29, 2025) were fused together and the probability of the estimated magnitude values were presented. The results show that DS was able to estimate the magnitude of the upcoming earthquake with a desirable uncertainty.

Abstract Whether and when earthquakes of different sizes can be distinguished early in their rupture process is critical to improving earthquake early warning (EEW) systems. To address this question, we develop a transformer‐based framework called source time function Magnitude Network (STF‐MgNet), which leverages earthquake STFs with M w 5.5 to investigate whether the initial stages of the rupture process can predict the earthquake's final magnitudes. The proposed STF‐MgNet utilizes transformer blocks in conjunction with a U‐Net architecture to effectively manage long‐range dependencies and capture essential information in data sets, boosting its performance. Training on 2,126 global M w ≥ 5.5 earthquakes reveals three distinct nucleation‐phase diagnostic regimes: (a) For 5.5 M w ≤ 7 events, subsecond analysis of the STF achieves 80.0% magnitude estimation accuracy; (b) For 7 < M w ≤ 8 earthquakes, 3–5 s of observations capture key features of STFs, yielding 86.9% accuracy; (c) For M w > 8 events, approximately 5 s of monitoring resolves the interplay between determinism and stochasticity, maintaining 83.7% accuracy despite greater complexity. Thus, no matter how an earthquake begins, earthquakes of different sizes can be distinguished at some point. Moreover, the earthquake magnitude prediction accuracy of STF‐MgNet improves with increasing input sequences of STFs. This study supports the hypothesis that rupture onset at nucleation differs for earthquakes of different final magnitudes.

Analysis of AI models for earthquake magnitude prediction in Myanmar

Earthquakes are unpreventable natural disasters that result in many casualties and economic losses in the regions where they occur. Earthquake prediction and seismic risk assessments are essential in minimising these losses. Due to the complex nature of seismic events, it is necessary to use a cutting-edge methodology to predict earthquake occurrence effectively. Machine learning methods have been among the most efficient and current methods for solving complex nonlinear problems and analysing big datasets. Because of this feature, they are widely used for predicting earthquakes and earthquake parameters. This study focuses on applying machine learning methods to analyse seismic events in Western Turkey from 1975 to 2024. The aim is to compare the effectiveness of five machine learning approaches for predicting earthquake magnitudes: Long Short-Term Memory (LSTM), Adaptive Neuro-Fuzzy Inference Systems (ANFIS), Decision Tree (DT), Random Forest (RF), and Convolutional Neural Network (CNN). The outcomes of these applied methods are encouraging in terms of the prediction of magnitude. Among all the results, the LSTM method is slightly more successful than the other methods, with a Root Mean Square Error (RMSE) of 0.1391, Mean Square Error (MSE) of 0.0193, Mean Absolute Error (MAE) of 0.1046 and Mean Absolute Percentage Error (MAPE) of 3.0631%, respectively.

An improved genetic algorithm and clone selection optimization-based gated recurrent unit networks for earthquake magnitude prediction

Abstract Indonesia is situated within the Pacific Ring of Fire, making it highly susceptible to frequent seismic activities. High-magnitude earthquakes can result in substantial damage to communities, and earthquake magnitude prediction remains a highly complex field. However, with advancing technology, numerous studies have been conducted to predict earthquake magnitudes, particularly in Indonesia. Through the implementation of machine learning, several algorithmic models have demonstrated potential in earthquake prediction, specifically Random Forest and Multilayer Perceptron (MLP). This research aims to analyze the comparative performance of Random Forest and Multilayer Perceptron algorithms, along with the implementation of hyperparameter tuning for earthquake magnitude prediction in Indonesia. The study seeks to identify the most accurate algorithmic model for predicting earthquake magnitudes in Indonesia and to determine whether hyperparameter tuning enhances the performance of Random Forest and MLP algorithms. The features utilized in this study include latitude, longitude, depth, magnitude, and Julian’s day. Earthquake data was obtained from the United States Geological Survey (USGS) Earthquake website. Through the application of these features and hyperparameter tuning, this research identified Random Forest as the optimal model for earthquake magnitude prediction in Indonesia, achieving evaluation metrics of MSE at 0.1488, MAE at 0.2832, MAPE at 6.2154, and RMSE at 0.1384.

This study evaluates the performance of machine learning and hybrid deep learning models for predicting earthquake magnitudes using historical seismic data. Five models, including Random Forest (RF), ARIMA, Long Short-Term Memory (LSTM), CNN+LSTM, and Transformer + Gaussian Processes (GP), were compared using metrics such as Root Mean Squared Error (RMSE) and R2. The RF model was quite efficient, with an RMSE of 0.072 and an R2 of 0.30. However, it did not incorporate temporal analysis. ARIMA was also better, with an RMSE of 0.065 and R2 of 0.42, which is best suited for linear relationships. LSTM identified the sequential relations well and provided an RMSE of 0.097 and R2 of 0.51. The hybrid CNN+LSTM model outperformed standalone approaches with an RMSE of 0.090 and R2 of 0.58 by combining spatial and temporal features. The Transformer + GP model achieved the highest accuracy, with an RMSE of 0.063 and R2 of 0.62, offering robust uncertainty quantification through confidence intervals. These results highlight the superiority of hybrid models in seismic forecasting, demonstrating their potential to improve predictive accuracy and support better risk management strategies.

Earthquake magnitude prediction is critical for natural calamity prevention and mitigation, significantly reducing casualties and economic losses through timely warnings. This study introduces a novel approach by using spatio-temporal data from seismic records obtained from the Indian government seismology department and weather data sourced via VisualCrossing to predict earthquake magnitudes. By integrating environmental and seismic variables, the study explores their interrelationships to enhance predictive capabilities. The proposed framework incorporates a machine learning operations (MLOps)-driven pipeline using MLflow for automated data ingestion, preprocessing, model versioning, tracking, and deployment. This novel integration ensures adaptability to evolving datasets and facilitates dynamic model selection for optimal performance. Multiple machine learning algorithms, including Gradient Boosting, Light Gradient Boosting Machine (LightGBM), XGBoost, and Random Forest, are evaluated on dataset sizes of 20%, 35%, 65%, and 100%, with performance metrics such as Mean Absolute Error, Mean Squared Error, Root Mean Squared Error, and R2. The results reveal that Gradient Boosting performs optimally on smaller datasets, while LightGBM demonstrates superior accuracy with larger datasets, showcasing the pipeline’s flexibility and scalability. This research presents a scalable, robust, and resilient solution for earthquake magnitude prediction by combining diverse data sources with a dynamic and operational MLOps framework. The outcomes illustrate the potential of integrating advanced machine learning techniques with lifecycle management practices to enhance prediction accuracy and applicability in real-world seismic scenarios.

Earthquake prediction is one of the most challenging and vital tasks that demands new methodologies for improving the accuracy of predictions. The research aims to present how radon gas concentration fluctuations are associated with the prediction of earthquakes in the Eurasian-Indo-Australian Plates. The paper discusses a statistical method of forecasting earthquake magnitudes greater than M4.5 from real-time radon gas monitoring close to the Grindulu Fault, Pacitan, East Java, Indonesia. This developed model has had the least errors in the form of mean absolute error (MAE), 0.30; mean absolute percentage error (MAPE), 0.06; root mean square error (RMSE), 0.55; mean squared error (MSE), 0.30; symmetric mean absolute percentage error (SMAPE), 0.06; complex normalized mean absolute percentage error (cnMAPE), 0.97; error absolute average (EAA), 0.30; and error relative average (ERA), -0.11, showing great accuracy and uniformity in prediction. These observations support the model’s efficiency that may be adopted in earthquake early warning systems for better disaster preparedness. Predictive errors are reduced, and there is support for improved disaster management strategy, public safety education, and effective emergency response personnel training. This study can be used as a foothold for further advances in earthquake prediction methodologies and refinement of early warning systems.

Real-time earthquake magnitude prediction using designed machine learning ensemble trained on real and CTGAN generated synthetic data

This study conducts a comparative analysis of four machine learning algorithms—Decision Tree, Random Forest, Support Vector Machine (SVM), and Neural Network—to predict earthquake magnitudes using the United States Geological Survey (USGS) earthquake dataset. The analysis evaluates each model's performance based on key metrics: Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), and the coefficient of determination (R²). The Random Forest model demonstrated superior performance, achieving the lowest MAE (0.217051), lowest RMSE (0.322398), and highest R² (0.574261), indicating its robustness in capturing complex, non-linear relationships in seismic data. SVM also showed strong performance, with competitive accuracy and robustness. Decision Tree and Neural Network models, while useful, had comparatively higher error rates and lower R² values. The study highlights the potential of ensemble learning and kernel methods in enhancing earthquake magnitude prediction accuracy. Practical implications of the findings include the integration of these models into early warning systems, urban planning, and the insurance industry for better risk assessment and management. Despite the promising results, the study acknowledges limitations such as reliance on historical data and the computational intensity of certain models. Future research is suggested to explore additional data sources, advanced machine learning techniques, and more efficient algorithms to further improve predictive capabilities. By providing a comprehensive evaluation of these models, this research contributes valuable insights into the effectiveness of various machine learning techniques for earthquake prediction, guiding future efforts to develop more accurate and reliable predictive models.

Explainable earthquake magnitude prediction with hybrid modeling and spatio-temporal data for scalability

DFTQuake: Tripartite Fourier attention and dendrite network for real-time early prediction of earthquake magnitude and peak ground acceleration

The application of machine learning in predicting earthquake magnitudes is crucial due to its ability to process extensive data sets and identify intricate patterns, thereby enhancing the accuracy and timeliness of predictions. This capability is essential for improving readiness and relief techniques against seismic activities. This study introduces a novel hybrid ensemble model, the HEM NAEMP, specifically evolved for predicting earthquake magnitudes along the North Anatolian Fault zone. The model integrates data from both the North Anatolian and San Andreas fault zones-the latter selected due to its tectonic similarity-to develop a comprehensive dataset that includes newly extracted features. The novelty of this study lies in the combination of data from two different fault lines to create a new dataset, the extraction of novel features, and the development of a previously unused model leveraging this dataset and its features. The HEM NAEMP model employs a several of regression algorithms, including k-nearest neighbors, random forest, support vector machine, decision tree and extreme gradient boosting, to effectively predict earthquake magnitude. The evaluation metrics for the model are as follows: mean squared error (MSE) of 0.011, mean absolute error (MAE) of 0.064, root mean squared error (RMSE) of 0.108, mean absolute percentage error (MAPE) of 0.268, R Square (R2) of 0.92 and training time of 2.44 sec. These results are compared against those from a Long-Short Term Memory (LSTM), Convolutional Neural Network (CNN) and AutoRegressive Integrated Moving Average (ARIMA) models, demonstrating that HEM NAEMP has mostly lower error rates in MAE and MAPE and high score in R2, as well as reduced training time, thereby confirming its viability and efficiency.

Turkey faces significant seismic risks, necessitating accurate earthquake forecasting for effective disaster preparedness. This study employs advanced Artificial Neural Networks (ANN) to predict earthquake magnitudes and assess risks specific to Turkey. Focusing on a distinct segment of the USGS earthquake catalogue from January 2014 to August 2023, the research tailors ANN algorithms with five layers to Turkey's seismic challenges. Rigorous dataset cleaning and processing ensure accuracy, with the ANN model demonstrating exceptional alignment with earthquake data ( RMSE: 0.078, R2: 0.89). Comparative evaluations highlight the effectiveness of ANN models in forecasting earthquake magnitudes in Turkey. The study explores the spatial distribution of earthquake risk across Turkey through an ANN-based map, emphasizing the critical window for preventive measures in this seismically active region. The analysis ensures further enhancement of model accuracy in seismic-prone areas globally. This study advances earthquake prediction by showcasing the high accuracy of our five-layer ANN model in forecasting magnitudes and spatial risk, significantly improving disaster preparedness and risk management in regions such as Turkey.