TELKOMNIKA Telecommunication Computing Electronics and Control

Vol. 23, No. 3, June 2025, pp. 825~836

ISSN: 1693-6930, DOI: 10.12928/TELKOMNIKA.v23i3.26494  825

Earthquake magnitude prediction based on radon cloud data

near Grindulu fault, Indonesia using the statistical method

Sunarno1, Thomas Oka Pratama1, Faridah1, Nugroho Ananto2, Hermin Kartika Sari3, Rony Wijaya1,4,
Memory Motivanisman Waruwu1,4
1
Department of Nuclear Engineering and Engineering Physics, Faculty of Engineering, Universitas Gadjah Mada, Yogyakarta, Indonesia
2
Sinergi, Jakarta, Indonesia
3
Department of Chemical Engineering, Politeknik Negeri Bandung, Bandung, Indonesia
4
Amakusa Instrumentation Technology, Yogyakarta, Indonesia

Article Info ABSTRACT

Article history: Earthquake prediction is one of the most challenging and vital tasks that
demands new methodologies for improving the accuracy of predictions. The
Received Aug 9, 2024 research aims to present how radon gas concentration fluctuations are
Revised Mar 6, 2025 associated with the prediction of earthquakes in the Eurasian-Indo-Australian
Accepted Mar 23, 2025 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
Keywords: 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),
Earthquake magnitude 0.55; mean squared error (MSE), 0.30; symmetric mean absolute percentage
Mitigation error (SMAPE), 0.06; complex normalized mean absolute percentage error
Prediction (cnMAPE), 0.97; error absolute average (EAA), 0.30; and error relative
Radon average (ERA), -0.11, showing great accuracy and uniformity in prediction.
Statistic 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.
This is an open access article under the CC BY-SA license.

Corresponding Author:
Thomas Oka Pratama
Department of Nuclear Engineering and Engineering Physics, Faculty of Engineering
Universitas Gadjah Mada
St. Grafika No. 2, Sinduadi, Mlati, Sleman, Yogyakarta, Indonesia
Email: [email protected]

1. INTRODUCTION
Earthquakes, as one of the most devastating natural disasters, strike suddenly, inflicting widespread
damage to structures and tragically claiming lives [1], [2]. Numerous research efforts have explored the
potential for earthquake prediction through the examination of various precursory signals, including animal
behaviour, temperature fluctuations, radon gas emissions, seismic activity alterations, and related indicators
[3], [4]. However, due to the inconsistent presence of these precursory indicators before each seismic event,
the standardization of forecasting approaches remains a significant challenge [5]. Inaccurate forecasts in many
existing methods contribute to the devastating consequences of earthquakes [2]. Radon gas has long been

Journal homepage: http://journal.uad.ac.id/index.php/TELKOMNIKA

826  ISSN: 1693-6930

considered a potential early indicator of earthquakes [6]–[12]. Yet a comprehensive prediction system
specifying the date, time, magnitude, and precise location of impending earthquakes remains [2], [9]–[12].
Recently, artificial intelligence (AI)-driven approaches have shown promising advancements in
earthquake prediction, surpassing traditional methods in accuracy. These techniques leverage machine learning
to analyze diverse data sources, including animal behaviour, weather patterns, groundwater levels, chemical
changes, and seismic activity, enabling the identification of potential warning signs before earthquakes occur.
AI-based methodologies are currently helpful for earthquake prediction, emphasizing its potential to improve
preparedness and response strategies in at-risk areas [2], [13]–[18]. While short-term earthquake prediction
with specific magnitude and location remains a challenge, this study presents a unique approach for predicting
earthquakes with a magnitude above M4.5 in Indonesia between the Eurasian and Indo-Australian Plates [19].
Zhang’s study constructs four models to analyse the mechanisms of radon variation under natural and
seismic conditions using the extreme gradient boosting method. Analysis of the precursory mechanisms of
these radon anomalies found that radon anomalies are most likely caused by increases in radon emanation due
to the earthquake-induced formation of microfractures in rock [20]. Related works in the field of earthquake
prediction reveal diverse approaches and methodologies, each with its unique set of challenges and potential.
Walia’s assessment highlights the absence of a definitive model linking earthquakes and radon anomalies,
underscoring the ongoing need for further validation of proposed models [21]. On the other hand, the
application of belief rule-based expert system (BRBES) demonstrates the potential to anticipate earthquake
occurrences within a maximum of 12 hours, drawing upon data on animal behaviour, environmental shifts, and
chemical variations [13]. Contradicting the conventional four-stage prediction framework, research on the
Haicheng earthquake indicates that the observed rise in seismic activity preceding the event does not align with
the anticipated prediction stages [22]. Examining the seismic cycle based on historical data, an expert system
showcased the capability to detect 100% of earthquakes within 12 hours within specific parameters of range,
depth, and location, estimating magnitudes ranging from M3.6 to M9.1 across one-quarter of the Earth’s
surface [14]. Leveraging climate data, Hajikhodaverdikhan et al. [15] successfully predicts earthquakes near
Tabriz, Iran, boasting high accuracy and precision rates in monthly earthquake forecasts.
In the realm of earthquake prediction methodologies, machine learning and deep learning have
emerged as focal points, as evidenced by the application of various techniques such as pattern recognition
neural networks, recurrent neural networks, random forests, and linear programming boost ensemble
classifiers. These techniques have been separately employed to model the relationships between calculated
seismic parameters and forthcoming earthquakes with magnitudes greater than or equal to 5.5 within a one-
month prediction timeframe, though significant challenges persist in the integration of these models into
effective forecasting systems [16]. With the aid of the support vector regression method, climate data is utilized
to predict earthquake magnitudes in specific regions, achieving a precision rate of 96% for mean magnitude
forecasts and a high accuracy rate of 78% in projected monthly earthquake counts [15]. Conversely, the
application of linear regression in earthquake prediction through data mining, considering groundwater levels,
chemical changes, and radon gas in groundwater, has faced challenges in understanding the intricate interplay
of these factors without rigorous empirical investigation [18].
Reflecting on the utility of AI in earthquake prediction, Banna’s discussion emphasizes the effective
forecasting of earthquakes within specific magnitude ranges (M3 to M5), with limitations observed in
predicting high-magnitude events due to their relative rarity and unpredictable occurrence patterns. Notably,
significant errors in time and location prediction have been encountered, with deviations of up to 70 miles and
substantial variation in prediction time frames ranging from 20 days to 5 months [2]. Based on Tehseen et al.’s
study [19], the accuracy proposed expert system for making earthquake predictions using an independent test
set is shown in Table 1.

Table 1. Accuracy is claimed in an expert system using an independent test set [19]
References Number of earthquake records Accuracy (%) Magnitude range
[23] 9531 69.8 ≥2.0
[24] 12690 50.14 ≥3.0
[25] 337 63 ≥3.0
[26] 10567 40 0.1–5.9

These findings underscore the complexity and inherent uncertainties associated with earthquake
prediction, driving the ongoing exploration and refinement of methodologies to enhance predictive accuracy
and reliability. The research trend has shifted towards machine learning and deep learning methodologies since
2018, marking a significant transition in earthquake prediction techniques [19]. Previous research has focused
on radon gas concentration fluctuations one to four days before an earthquake event based on Thomas Oka’s

TELKOMNIKA Telecommun Comput El Control, Vol. 23, No. 3, June 2025: 825-836
TELKOMNIKA Telecommun Comput El Control  827

earthquake date prediction method between the Eurasia and Indo-Australia plates in Indonesia [27]. The latest
research on predicting earthquake magnitudes using a linear regression technique has achieved the lowest
values across various evaluation metrics based on the radon gas fluctuation in Yogyakarta, Indonesia. The
results include a standard deviation of 0.40, mean absolute error (MAE) of 0.30, mean absolute percentage
error (MAPE) of 6%, root mean square error (RMSE) of 0.52, mean squared error (MSE) of 0.28, symmetric
mean absolute percentage error (SMAPE) of 0.06, and cnSMAPE of 0.97.
There have been one to four days of earthquake date prediction before the event based on radon gas
concentration measurements near the Grindulu Fault in Pacitan, East Java, Indonesia. Still, the magnitude is
not yet predicted [27]. This research seeks magnitude prediction through radon gas concentration fluctuations
in the Eurasian and Indo-Australian Plates, focusing on earthquakes with magnitudes greater than M4.5, and
then implementing an earthquake early warning system based on the radon gas concentration. This study
discusses using the statistical method for magnitude prediction to improve the earthquake early warning
system, to reduce the risk of being affected by disasters, and to prepare emergency response actions.

2. METHOD
The radon gas telemonitoring system is installed near the active Grindulu fault in Pacitan, East Java,
Indonesia, in such a way that the system becomes most sensitive to earthquakes. The radon gas sensor is exactly
put in a controlled chamber for optimally achieving accurate data, which is approximately around the fault
area. The frequency of radon gas measurements is adjusted every 10 minutes to minimize the impact of
radiation emissions from Actinium and Thoron [28]. Figure 1 illustrates the overall architecture of the
monitoring system. The sensor readings are logged and sent continuously to an ESP32 microprocessor and
then to a cloud server, enabling real-time monitoring provided by an internet connection. The readings of the
concentration of the radon gas are logged securely to a dedicated storage server, from which the users can
retrieve them using a cloud-based interface. Besides, earthquake data analysis relies on safely stored radon
measurements and earthquake occurrence records to allow reliable monitoring and prediction capabilities.
The earthquake magnitude prediction algorithm is developed using statistical methods, employing
data from Radon clouds and earthquake occurrences. Performance evaluation of the model includes metrics
such as MAE, MAPE, RMSE, MSE, SMAPE, and cnSMAPE. The optimal model is selected for deployment
on a cloud server for earthquake prediction notifications.

Figure 1. Earthquake magnitude prediction scheme [29]

Table 2 presents the dataset structure for radon gas concentration, arranged based on the methodology
developed by Pratama et al. [27]. Subsequently, detailed information regarding radon gas concentration and
earthquake activities was structured and tabulated systematically in Table 3 following Thomas Oka’s
framework. The training dataset consists of radon gas concentration data recorded during the prediction interval
of earthquake activities. Data were collected from 156 cases in the study area, of which 80% were utilized as
training data and 20% for predicting and testing earthquake magnitudes. Data structure and distribution align
closely with the method already established, which is standard and reliable for outcomes.
The algorithm for earthquake magnitude prediction is determined through stages starting from data
tabulation and synchronization until the performance evaluation values are obtained. The algorithm for each

Earthquake magnitude prediction based on radon cloud data near Grindulu fault, Indonesia … (Sunarno)
828  ISSN: 1693-6930

station is determined by the flowchart shown in Figure 2. Subsequently, the data is computed to determine the
a value. A selection occurs when an earthquake date prediction is made using the existing algorithm. Various
combinations of a are tested to derive b, the characteristic coefficient. The next step is to establish the
relationship between b and c, where c represents the earthquake magnitude recorded by the earthquake
precursor telemonitoring station. This process results in a polynomial formula that defines the relationship
between b and c. The earthquake magnitude can be predicted by substituting b into this polynomial formula.
Following this, the performance evaluation of the predicted earthquake magnitudes is calculated. The
evaluation of the machine learning process encompassed metrics such as MAE, MAPE, RMSE, MSE, SMAPE,
and cnSMAPE.

Table 2. Data set composition [27]

Variable Description
x The day when the algorithm prediction was completed based on the method of Thomas Oka for Pacitan station [27].
Rx Radon average day x
R(x-1) Radon average day x-1
R(x-2) Radon average day x-2
R(x-6) Radon average day x-6
R(x-7) Radon average day x-7
HR(x-3) Radon average 3 days before R(x-2) = average R(x-3) to R(x-5)
HR(x-7) Radon average 7 days before R(x-2) = average R(x-3) to R(x-9)
HR(x-14) Radon average 14 days before R(x-2) = average R(x-3) to R(x-17)

Table 3. Example of dataset

Earthquake
HR HR HR R R R R R R R Earthquake Real
date
(x-14) (x-7) (x-3) (x-7) (x-6) (x-5) (x-4) (x-3) (x-2) (x-1) date magnitude
prediction
12/2/2023 87.17 91.50 102.86 83.23 74.81 91.70 105.12 111.76 104.13 122.15 12/3/2023 4.7
12/7/2023 100.48 115.76 122.38 104.13 122.15 143.17 118.75 105.22 118.59 86.00 12/11/2023 4.9
12/18/2023 99.86 92.67 104.19 92.32 83.15 102.89 103.07 106.59 101.28 51.95 12/22/2023 4.8
12/25/2023 81.82 70.96 77.29 52.73 58.91 83.57 71.01 77.29 64.34 75.01 12/26/2023 4.6
12/26/2023 81.34 65.69 70.88 58.91 83.57 71.01 77.29 64.34 75.01 69.46 12/28/2023 5.1

Figure 2. Magnitude prediction algorithm determination flowchart

TELKOMNIKA Telecommun Comput El Control, Vol. 23, No. 3, June 2025: 825-836
TELKOMNIKA Telecommun Comput El Control  829

Algorithms that have completed the determination stage are subsequently tested. The testing phase
occurred from April 1, 2022, to May 30, 2024, at the Pacitan radon gas concentration telemonitoring station. This
testing aims to identify the best algorithm according to the specified criteria. The best algorithm is then
implemented in the server cloud and sends the notification to Telegram via the Telegram API. Thus, earthquake
magnitude predictions can be automatically made based on the designed algorithm using radon gas concentration
measurements for one to four days later between the Eurasia and Indo-Australia Plates with a magnitude above M4.5.

3. RESULTS AND DISCUSSION

In this section, we report the earthquake magnitude prediction when there is an alert from Thomas
Oka’s earthquake date prediction for the radon gas telemonitoring station in Pacitan [27]. When the earthquake
date prediction alarm is active, data on Radon gas concentration is collected for magnitude prediction
calculation. Based on the data processing results conducted according to the research flow in Figure 2, three
gains were selected: Gain A (|R(x-7) – HR(x-7)|), Gain B (|R(x-3)-HR(x-7)|), and Gain C (|R(x-7)-R(x-1)|).
The relationship between these gains and the actual earthquake magnitude was determined using polynomial
equations. In this study, three polynomial degrees were used: 6th degree (Gain 1), 5th degree (Gain 2), and 4th
degree (Gain 3) for each Gain (A, B, and C). This results in Gain 1A, Gain 2A, Gain 3A, Gain 1B, Gain 2B,
Gain 3B, Gain 1C, Gain 2C, and Gain 3C. Predicted earthquake magnitude values were generated by
substituting the gains into the equation. These predictions were then evaluated for both training and test data.
Gains 3A, 1B, and 1C were excluded from the summary because they produced poor results, with predictions
exceeding M10, a value that is not possible for earthquake magnitude scales.
The performance evaluation results for the training data are shown in Table 4. From this Table 4, it is
evident that Gain 2B has the lowest evaluation performance for the MAE (0.37), MAPE (0.07), RMSE (0.61),
MSE (0.37), and SMAPE (0.07) parameters, while its complex normalized mean absolute percentage error
(cnMAPE) value is the highest (0.96), the same as Gain 3B. Additionally, it has the lowest absolute standard
deviation (0.35) and relative standard deviation (0.51) compared to the other gains. Therefore, Gain 2B is the
gain with the best evaluation for the training data compared to the other gains. The equation for Gain 2B is
shown in (1), where x is the gain.
Table 5 presents the performance evaluation results of the equations used to predict earthquake
magnitude using test data. From this Table 5, it can be seen that Gain 3B has the lowest evaluation values for
MAE (0.30), MAPE (0.06), RMSE (0.55), MSE (0.30), and SMAPE (0.06), while its cnMAPE value is the highest
(0.97), the same as Gain 2C. Additionally, it has the lowest absolute standard deviation (0.28) and relative standard
deviation (0.40) compared to the other gains. Therefore, Gain 3B is the gain with the best evaluation for the test
data compared to the other gains. The equation for Gain 3B is shown in (2), with x is Gain 3B.
The MAE, MAPE, RMSE, MSE, SMAPE, error absolute average (EAA), and error relative average
(ERA) values for earthquake magnitude prediction using the statistical method are close to 0, and cnSMAPE
is close to 1. This indicates that the earthquake magnitude prediction model based on radon gas concentration
telemonitoring data is acceptable. The model demonstrates good prediction performance and can be considered
reliable for practical applications in earthquake prediction.

Mpred, Gain 2B = -1E-09x5 - 3E-07x 4 + 5E-05x3 - 0.0022x2 + 0.0276x + 5.1 (1)

Mpred, Gain 3B = -2E-09x5 + 7E-07x4 - 7E-05x3 + 0.0034x2 - 0.0665x + 5.5134 (2)

Table 4. Performance evaluation of data training set

MAE MAPE RMSE MSE SMAPE cnSMAPE Stdev absolut Stdev relatives
Gain 1 A 0.64 0.12 1.17 1.37 0.15 0.92 0.90 1.03
Gain 2 A 0.54 0.10 0.73 0.54 0.13 0.93 0.81 0.93
Gain 2 B 0.37 0.07 0.61 0.37 0.07 0.96 0.35 0.51
Gain 3 B 0.42 0.08 0.65 0.42 0.08 0.96 0.37 0.55
Gain 2 C 0.88 0.17 0.94 0.88 0.12 0.94 2.12 2.22
Gain 3 C 0.62 0.12 0.79 0.62 0.17 0.92 0.95 1.13

Table 5. Earthquake magnitude prediction data test error evaluation

Error index MAE (s) MAPE RMSE MSE SMAPE cnSMAPE EAA ERA
Gain 1 A 0.81 0.15 0.90 0.81 0.33 0.84 0.81 0.61
Gain 2 A 0.50 0.10 0.71 0.50 0.11 0.94 0.50 0.29
Gain 2 B 0.40 0.08 0.63 0.40 0.10 0.95 0.40 0.10
Gain 3 B 0.30 0.06 0.55 0.30 0.06 0.97 0.30 -0.11
Gain 2 C 0.35 0.07 0.59 0.35 0.07 0.97 0.35 -0.16
Gain 3 C 0.36 0.07 0.60 0.36 0.07 0.97 0.36 -0.17
Earthquake magnitude prediction based on radon cloud data near Grindulu fault, Indonesia … (Sunarno)
830  ISSN: 1693-6930

We analyzed the error value during the model evaluation process, representing the disparity between
the actual and predicted magnitudes. Figure 3 presents the boxplot of the error evaluation for the data test.
Magnitude predictions using Gain 1A and Gain 2A models have a wide error distribution. All models have
outlier errors visible at points outside the quartile limits. In Gain 1A and Gain 2A, there are several significant
outliers, indicating that in some cases, the errors can be very large. The models with Gain 2B, Gain 3B, Gain
2C, and Gain 3C have error distributions for the first quartile (Q1) around 0.1-0.2 while for the third quartile
(Q3) around 0.3-0.4. The model using these gains has a relatively narrow IQR, indicating small, consistent,
and controlled errors. The box plot further strengthens the model’s ability to predict earthquake magnitude
with a relatively low error range.
Furthermore, to analyze the signs of deviations produced by regressors in their predictions, Figure 4
shows histograms of data test errors for every model. A positive error indicates that the prediction was lower
than the actual value, while a negative error indicates that the prediction was higher than the actual value. The
method prediction shows a higher quantity of negative errors than positive ones (negative bias), as can be
noticed in the histograms of Figure 4. Negative bias in the error frequency graph refers to the tendency of errors
to lean towards lower values than the actual magnitude values. In the error frequency graph context, negative
bias is reflected in a distribution of errors that tend to be too low or negative.
All models have the highest error frequency at -0.25. The Gain1A Figure 4(a) and Gain 2A error
distributions tend to be centered around -0.50 to 0.00, with a peak at -0.25, but for Gain 2A, it is slightly more
spread-out Figure 4(b). The Gain 2B Figure 4(c) and Gain 3B Figure 4(d) distributions are more centered, with
a frequency peak at -0.25 with a frequency of 10. Gain 2 B and Gain 3 B have the narrowest error distributions
with a frequency peak at -0.25 and a smaller error range, indicating more consistent predictions. Gain 2C
Figure 4(e) and Gain 3C Figure 4(f) showed a wider error range, with a more even distribution across the error
range, indicating that these predictions tend to be more variable and less consistent than Gain 2B and Gain 3B.
Overall, Gain 2B and 3B show more stable and consistent performance than the other gain configurations, with
smaller errors and a more centered distribution. This is in line with the results of the data test prediction
performance evaluation in Table 6.

Figure 3. Data test boxplot error

Compared to other studies (Table 6) [13]-[17], [19], [29]-[49], magnitude predictions based on radon
gas cloud data are rare, mostly using seismic data. Machine learning is not always better than statistical methods
because machine learning methods use statistical calculations processed by machines/systems. The results of
earthquake magnitude prediction show a comparison of earthquake magnitude prediction accuracy with
differences in the main error metrics. The results of this research have a lower MAE and MAPE (0.30 and
0.06), indicating a smaller average absolute error and a more accurate percentage error than the research results
using machine learning (MAE 0.33 and MAPE 6.03%). However, the results had lower RMSE and MSE (0.51
and 0.26), indicating smaller overall squared errors and more consistent predictions. The magnitude prediction
results have identical SMAPE and cnSMAPE values (0.06 and 0.97), indicating similar performance in
symmetric percentage error. If the focus is on absolute error, the results of this research are superior. In contrast,
if the squared error is emphasized, the research results using machine learning methods are more optimal [42].

TELKOMNIKA Telecommun Comput El Control, Vol. 23, No. 3, June 2025: 825-836
TELKOMNIKA Telecommun Comput El Control  831

To advance earthquake prediction efforts, it is essential to develop new algorithms tailored to

predicting earthquake magnitudes in various locations, as radon gas concentration characteristics differ by
region. While challenges persist in forecasting rare high-magnitude earthquakes, this study marks a significant
step forward. Future research should focus on integrating additional data to reduce prediction error, especially
for magnitudes above M6 with rare events Figure 5, and for predicting the specific area of earthquake prediction
in Indonesia, enriching the understanding of earthquake prediction and emphasizing improving methodologies
in Indonesia.

(a) (b)

(c) (d)

(e) (f)

Figure 4. Error frequency of (a) Gain 1A, (b) Gain 2A, (c) Gain 2B, (d) Gain3B, (e) Gain 2C, and (f) Gain 3C

Figure 5. Earthquake data frequency

Earthquake magnitude prediction based on radon cloud data near Grindulu fault, Indonesia … (Sunarno)
832  ISSN: 1693-6930

Table 6. Previous research results

Ref. Key predictors Methods used Prediction scope Performance highlights
[13] Animal, BRBES 12 hours, Magnitude >6.5 Superior prediction (AUC=0.969),
environmental, and outperforming FLBES and ANN
chemical indicators
[19] Seismic data Expert system Varied magnitude Limited accuracy (<70%)
(M0.1–M5.9)
[14] Seismic activity Expert system Global coverage, 12-hour Exceptional accuracy (100%)
interval, M3.6–M9.1
[15] Meteorological, Support vector Monthly forecast Precision (96% magnitude), moderate
seismic data regression accuracy (78% count)
[16] Seismic parameters Neural networks, Monthly, Hindukush region, Moderate accuracy: Training (79%),
Random Forest M≥5.5 Testing (65%)
[17] Big data seismic Regression models Weekly, California, M3–M7 Low errors across magnitudes (MSE
parameters <0.79, MAE<0.59)
[30] Seismic Multilayer perceptron Short-term, M>4 classification Moderate accuracy (73.8%)
measurements
[31] Seismic electric Artificial neural Days, Greece, M≥5.2 Good overall accuracy (84%),
signals (SES) networks reduced for high magnitudes (58%)
[32] Seismic variables Adaptive neuro-fuzzy Iran, M>5.5 Strong model performance (R²=0.94,
system (ANFIS) RMSE=0.173)
[33] Seismic data Probabilistic neural Monthly, California, multi- Reliable (R score: 0.62–0.78)
network class magnitudes
[34] Historical seismic Neural network Monthly, Taiwan, M≥6 Low predictive capability (R
events score=0.303)
[35] Seismic data Classifier ensembles 5 days, Chile, M4–M7 Varied sensitivity (46–90%), excellent
specificity (89–100%)
[36] Seismic data SVM, Naïve Bayes Daily, Indonesia Moderate prediction errors
(RMSE=0.751, MAE=0.598)
[29] Radon time-series ML techniques Short-term (1–4 days), Java, XGBoost best (MAE=0.33,
data (XGBoost, SVM, RF) M>4.5 RMSE=0.51)
[37] Turkish seismic LSTM, CNN, ARIMA 36-month, Turkey LSTM most effective for long-term
data magnitude prediction
[38] Climate, seismic LSTM, Transformer Japan, Indonesia, Himalaya, High accuracy (MAE=0.066,
data magnitude frequency MSE=0.007)
[39] Geospatial seismic Random forest Turkey, Himalayan region Moderate accuracy (62.9%), balanced
features classifier precision and recall
[40] Geomorphological MLP neural network Vietnam fault region, Strong predictions (R²≈0.87,
indicators maximum magnitude RMSE≈0.10)
[41] Magnitude, spatial VMD-BP neural Tibet, Yunnan, M>4 High accuracy (R²≈0.93), superior to
coordinates network traditional BP
[42] Seismicity hDCA, SVM, KNN, Monthly, Sichuan, China, hDCA excellent (precision=0.73,
parameters BPNN M>4.5 AUC=0.97)
[43] Fault density, depth, Deep neural networks, Weekly, Iran-Himalayan Excellent specificity (≥90%), strong
spatial features SVM region, multi-class magnitudes precision (81–88%)
[44] Global seismic data K-means enhanced Weekly, Global, M≥5.5 High positive predictive value (96%)
ANN
[45] Seismic risk ML models (RF, Turkey earthquake risk ANN very effective (MAE=0.176,
parameters XGBoost, ANN) mapping RMSE=0.181, R²≈1)
[46] Seismic structure ANN, Tabu-search General earthquake design Low errors (MAE=0.081,
parameters parameters RMSE=0.116)
[47] Electromagnetic, CNN with 3D features Multi-class magnitude Excellent precision (97%), recall
vibration signals classification (98%)
[48] Seismic event data CNN-BiLSTM hybrid Japan, China, Magnitude 4.5–6 Moderate overall accuracy (69%),
varying sensitivity
[49] Bangladesh seismic Logistic regression, Bangladesh, Magnitude ≥5.0 Logistic regression superior (≈89%
indicators SVM accuracy)

4. CONCLUSION
The research findings confirmed the efficacy of the statistical method in forecasting earthquake
magnitudes using radon cloud data with Gain 3B, achieving the lowest error values across several evaluation
metrics: MAE (0.30), MAPE (0.06), RMSE (0.55), MSE (0.30), SMAPE (0.06), cnMAPE (0.97), EAA (0.3),
and ERA (-0.11), demonstrating smaller errors and more centralized distribution. These findings confirm the
implementation of the statistical model in the server cloud of the earthquake early warning system, which can
provide more accurate and timely predictions above M4.5. By improving the accuracy of prediction, this model
enhances emergency preparedness, supports public education on safety measures, and equips responders with
tools to manage earthquake-related disasters. The findings of this research are a very strong foundation for
further developments in the field of refinement of forecasting models and strengthening early warning systems
to mitigate seismic events.

TELKOMNIKA Telecommun Comput El Control, Vol. 23, No. 3, June 2025: 825-836
TELKOMNIKA Telecommun Comput El Control  833

ACKNOWLEDGMENTS
The author deeply thanks the Sensor System and Tele-Control Laboratory team at the Department of
Nuclear Engineering and Engineering Physics, Universitas Gadjah Mada, Indonesia, for providing the radon
gas telemonitoring system. We also thank Postdam Geofon and the Indonesian Meteorology, Climatology, and
Geophysics Agency (Badan Meteorologi, Klimatologi, dan Geofisika) for providing the earthquake data.

FUNDING INFORMATION
We gratefully acknowledge the financial support from the National Research and Innovation Agency of
the Republic of Indonesia (Badan Riset dan Inovasi Nasional) and the Indonesia Endowment Fund for Education
(Lembaga Pengelola Dana Pendidikan) through the Research and Innovation for Advanced Indonesia Scheme
Wave I with contract number 34/IV/KS/06/2022 and 2439/UN1/DITLIT/DitLit/KP.01.03/2022 for the 2023/2024
budget year, which facilitated the completion of this research.

AUTHOR CONTRIBUTIONS STATEMENT

This journal uses the Contributor Roles Taxonomy (CRediT) to recognize individual author
contributions, reduce authorship disputes, and facilitate collaboration.

Name of Author C M So Va Fo I R D O E Vi Su P Fu
Sunarno ✓ ✓ ✓ ✓ ✓ ✓ ✓
Thomas Oka Pratama ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓
Faridah ✓ ✓ ✓ ✓
Nugroho Ananto ✓ ✓ ✓ ✓ ✓
Hermin Kartika Sari ✓ ✓ ✓ ✓ ✓ ✓ ✓
Rony Wijaya ✓ ✓ ✓ ✓ ✓
Memory Motivanisman Waruw ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓

C : Conceptualization I : Investigation Vi : Visualization

M : Methodology R : Resources Su : Supervision
So : Software D : Data Curation P : Project administration
Va : Validation O : Writing - Original Draft Fu : Funding acquisition
Fo : Formal analysis E : Writing - Review & Editing

CONFLICT OF INTEREST STATEMENT

The authors state no conflict of interest.

DATA AVAILABILITY
The data supporting this study’s findings are available on request from the corresponding author
[email protected]. The data, which contain information that could compromise the privacy of research
participants, are not publicly available due to certain restrictions.

REFERENCES
[1] M. Senthilkumar, D. Gnanasundar, B. Mohapatra, A. K. Jain, A. Nagar, and P. K. Parchure, “Earthquake prediction from high
frequency groundwater level data: A case study from Gujarat, India,” HydroResearch, vol. 3, pp. 118–123, 2020, doi:
10.1016/j.hydres.2020.10.004.
[2] M. H. Al Banna et al., “Application of artificial intelligence in predicting earthquakes: State-of-the-art and future challenges,” IEEE
Access, vol. 8, pp. 192880–192923, 2020, doi: 10.1109/ACCESS.2020.3029859.
[3] B. Bahrami and M. Shafiee, “Fuzzy descriptor models for earthquake time prediction using seismic time series,” International
Journal of Uncertainty, Fuzziness and Knowledge-Based Systems, vol. 23, no. 04, pp. 505–519, Aug. 2015, doi:
10.1142/S0218488515500221.
[4] A. Ahumada, A. Altunkaynak, and A. Ayoub, “Fuzzy logic-based attenuation relationships of strong motion earthquake records,”
Expert Systems with Applications, vol. 42, no. 3, pp. 1287–1297, Feb. 2015, doi: 10.1016/j.eswa.2014.09.035.
[5] S. Farahi Ghasre Aboonasr, A. Zamani, F. Razavipour, and R. Boostani, “Earthquake hazard assessment in the Zagros Orogenic
42, Belt of Iran using a fuzzy rule-based model,” Acta Geophysica, vol. 65, no. 4, pp. 589–605, Aug. 2017, doi: 10.1007/s11600-
017-0055-4.
[6] C. H. Scholz, L. R. Sykes, and Y. P. Aggarwal, “Earthquake prediction: A physical basis,” Science, vol. 181, no. 4102, pp. 803–
810, Aug. 1973, doi: 10.1126/science.181.4102.803.
[7] M. Al-Hilal and M. Mouty, “Radon monitoring for earthquake prediction on Al-Ghab fault of Syria,” Nuclear Geophysics, vol. 8,
no. 3, pp. 291–299, 1994.

Earthquake magnitude prediction based on radon cloud data near Grindulu fault, Indonesia … (Sunarno)
834  ISSN: 1693-6930

[8] A. Alam, N. Wang, G. Zhao, and A. Barkat, “Implication of radon monitoring for earthquake surveillance using statistical
techniques: A case study of wenchuan earthquake,” Geofluids, vol. 2020, pp. 1–14, Mar. 2020, doi: 10.1155/2020/2429165.
[9] R. Gas, “EPSC 330 presentation earthquake precursors: Radon gas,” sappho.eps.mcgill.ca, 2011.
http://sappho.eps.mcgill.ca/~olivia/EES/Presentations/Presentations-2018/Emily-Dazee_PresentationSummary.pdf (accessed Oct.
02, 2024).
[10] A. S. Bafti, “The relation of concentration radon gas in underground water with tectonic movements and Seismicity. Case study:
Jorjafk fault, Central Iran,” Scientific Quarterly Journal of Geosciences, vol. 30, no. 3, pp. 189–204, 2020.
[11] C. Lingling and G. Antai, “Correlativity between short-term change of fault gas radon concentration and air temperature &
atmospheric pressure,” Acta Seismologica Sinica, vol. 36, no. 4, pp. 719–729, 2014.
[12] C.-C. Fu et al., “Preseismic anomalies in soil-gas radon associated with 2016 M 6.6 Meinong earthquake, Southern Taiwan,”
Terrestrial, Atmospheric and Oceanic Sciences, vol. 28, no. 5, pp. 787–798, 2017, doi: 10.3319/TAO.2017.03.22.01.
[13] M. S. Hossain, A. Al Hasan, S. Guha, and K. Andersson, “A belief rule based expert system to predict earthquake under uncertainty,”
Journal of Wireless Mobile Networks, Ubiquitous Computing, and Dependable Applications, vol. 9, no. 2, pp. 26–41, 2018.
[14] A. Ikram and U. Qamar, “Developing an expert system based on association rules and predicate logic for earthquake prediction,”
Knowledge-Based Systems, vol. 75, pp. 87–103, Feb. 2015, doi: 10.1016/j.knosys.2014.11.024.
[15] P. Hajikhodaverdikhan, M. Nazari, M. Mohsenizadeh, S. Shamshirband, and K. Chau, “Earthquake prediction with meteorological
data by particle filter-based support vector regression,” Engineering Applications of Computational Fluid Mechanics, vol. 12, no.
1, pp. 679–688, Jan. 2018, doi: 10.1080/19942060.2018.1512010.
[16] K. M. Asim, F. M.-Álvarez, A. Basit, and T. Iqbal, “Earthquake magnitude prediction in Hindukush region using machine learning
techniques,” Natural Hazards, vol. 85, no. 1, pp. 471–486, Jan. 2017, doi: 10.1007/s11069-016-2579-3.
[17] G. A.–Cortés, A. M.–Esteban, X. Shang, and F. M.–Álvarez, “Earthquake prediction in California using regression algorithms and
cloud-based big data infrastructure,” Computers & Geosciences, vol. 115, pp. 198–210, Jun. 2018, doi:
10.1016/j.cageo.2017.10.011.
[18] T. Khan, M. Rabbani, S. M. T. Siddiquee, and A. Majumder, “An innovative data mining approach for determine earthquake
probability based on linear regression algorithm,” in 2019 IEEE International Conference on Electrical, Computer and
Communication Technologies (ICECCT), Feb. 2019, pp. 1–4. doi: 10.1109/ICECCT.2019.8869286.
[19] R. Tehseen, M. S. Farooq, and A. Abid, “Earthquake prediction using expert systems: A systematic mapping study,” Sustainability,
vol. 12, no. 6, Mar. 2020, doi: 10.3390/su12062420.
[20] S. Zhang, Z. Shi, G. Wang, R. Yan, and Z. Zhang, “Application of the extreme gradient boosting method to quantitatively analyze
the mechanism of radon anomalous change in Banglazhang hot spring before the Lijiang Mw 7.0 earthquake,” Journal of Hydrology,
vol. 612, Sep. 2022, doi: 10.1016/j.jhydrol.2022.128249.
[21] V. Walia, H. S. Virk, T. F. Yang, S. Mahajan, M. Walia, and B. S. Bajwa, “Earthquake prediction studies using radon as a precursor
in N-W Himalayas, India: A case study,” Terrestrial, Atmospheric and Oceanic Sciences, vol. 16, no. 4, pp. 775–795, 2005, doi:
10.3319/TAO.2005.16.4.775(GIG).
[22] K. Wang, “Predicting the 1975 haicheng earthquake,” Bulletin of the Seismological Society of America, vol. 96, no. 3, pp. 757–795,
Jun. 2006, doi: 10.1785/0120050191.
[23] M. Last, N. Rabinowitz, and G. Leonard, “Correction: Predicting the maximum earthquake magnitude from seismic data in Israel
and its neighboring countries,” PLOS ONE, vol. 11, no. 3, Mar. 2016, doi: 10.1371/journal.pone.0151751.
[24] S. Ghorbani, M. Barari, and M. Hoseini, “Presenting a new method to improve the detection of micro-seismic events,”
Environmental Monitoring and Assessment, vol. 190, no. 8, Aug. 2018, doi: 10.1007/s10661-018-6837-6.
[25] N. Tahernia, “Fuzzy-logic tree approach for seismic hazard analysis,” International Journal of Engineering and Technology, vol.
6, no. 3, pp. 182–185, 2014, doi: 10.7763/IJET.2014.V6.692.
[26] B. R.‐Leduc, C. Hulbert, N. Lubbers, K. Barros, C. J. Humphreys, and P. A. Johnson, “Machine learning predicts laboratory
earthquakes,” Geophysical Research Letters, vol. 44, no. 18, pp. 9276–9282, Sep. 2017, doi: 10.1002/2017GL074677.
[27] T. O. Pratama, S. Sunarno, M. M. Waruwu, and R. Wijaya, “Earthquake date prediction based on the fluctuation of radon gas
concentration near grundulu fault,” Jurnal Lingkungan dan Bencana Geologi, vol. 14, no. 2, pp. 104–113, 2023.
[28] H. Cember and T. E. Johnson, Introduction to health physics. 4th ed. New York: McGraw-Hill Medical, 2009.
[29] T. O. Pratama, S. Sunarno, A. B. Wijatna, and E. Haryono, “Cloud radon data for earthquake magnitude prediction using machine
learning,” IAES International Journal of Artificial Intelligence (IJ-AI), vol. 13, no. 4, pp. 4572–4582, Dec. 2024, doi:
10.11591/ijai.v13.i4.pp4572-4582.
[30] J. Mahmoudi, M. A. Arjomand, M. Rezaei, and M. H. Mohammadi, “Predicting the earthquake magnitude using the multilayer
perceptron neural network with two hidden layers,” Civil Engineering Journal, vol. 2, no. 1, pp. 1–12, Jan. 2016, doi: 10.28991/cej-
2016-00000008.
[31] M. Moustra, M. Avraamides, and C. Christodoulou, “Artificial neural networks for earthquake prediction using time series
magnitude data or Seismic Electric Signals,” Expert Systems with Applications, vol. 38, no. 12, pp. 15032–15039, Nov. 2011, doi:
10.1016/j.eswa.2011.05.043.
[32] M. Mirrashid, “Earthquake magnitude prediction by adaptive neuro-fuzzy inference system (ANFIS) based on fuzzy C-means
algorithm,” Natural Hazards, vol. 74, no. 3, pp. 1577–1593, Dec. 2014, doi: 10.1007/s11069-014-1264-7.
[33] H. Adeli and A. Panakkat, “A probabilistic neural network for earthquake magnitude prediction,” Neural Networks, vol. 22, no. 7,
pp. 1018–1024, Sep. 2009, doi: 10.1016/j.neunet.2009.05.003.
[34] J. Huang, X. Wang, Y. Zhao, C. Xin, and H. Xiang, “Large earthquake magnitude prediction in taiwan based on deep learning
neural network,” Neural Network World, vol. 28, no. 2, pp. 149–160, 2018, doi: 10.14311/NNW.2018.28.009.
[35] M. F.-Gómez, G. A.-Cortés, A. Troncoso, and F. M.-Álvarez, “Large earthquake magnitude prediction in chile with imbalanced
classifiers and ensemble learning,” Applied Sciences, vol. 7, no. 6, Jun. 2017, doi: 10.3390/app7060625.
[36] I. M. Murwantara, P. Yugopuspito, and R. Hermawan, “Comparison of machine learning performance for earthquake prediction in
Indonesia using 30 years historical data,” TELKOMNIKA (Telecommunication Computing Electronics and Control), vol. 18, no. 3,
pp. 1331–1342, Jun. 2020, doi: 10.12928/telkomnika.v18i3.14756.
[37] H. Ö. Çekim, H. N. Karakavak, G. Özel, and S. Tekin, “Earthquake magnitude prediction in Turkey: a comparative study of deep
learning methods, ARIMA and singular spectrum analysis,” Environmental Earth Sciences, vol. 82, no. 16, Aug. 2023, doi:
10.1007/s12665-023-11072-1.
[38] B. Sadhukhan, S. Chakraborty, S. Mukherjee, and R. K. Samanta, “Climatic and seismic data-driven deep learning model for
earthquake magnitude prediction,” Frontiers in Earth Science, vol. 11, Feb. 2023, doi: 10.3389/feart.2023.1082832.
[39] S. Sawantt et al., “Earthquake prognosis using machine learning,” ITM Web of Conferences, vol. 56, Aug. 2023, doi:
10.1051/itmconf/20235605017.

TELKOMNIKA Telecommun Comput El Control, Vol. 23, No. 3, June 2025: 825-836
TELKOMNIKA Telecommun Comput El Control  835

[40] C. D. Trong et al., “Using geomorphological indicators to predict earthquake magnitude (M Ob‒Max): A case study from cao bang
province and adjasent areas (Vietnam),” Geotectonics, vol. 56, no. 3, pp. 321–338, Jun. 2022, doi: 10.1134/S0016852122030104.
[41] J. Zhang and X. He, “Earthquake magnitude prediction using a VMD-BP neural network model,” Natural Hazards, vol. 117, no. 1,
pp. 189–205, May 2023, doi: 10.1007/s11069-023-05856-8.
[42] W. Zhou, H. Dong, and Y. Liang, “The deterministic dendritic cell algorithm with Haskell in earthquake magnitude prediction,”
Earth Science Informatics, vol. 13, no. 2, pp. 447–457, Jun. 2020, doi: 10.1007/s12145-020-00442-z.
[43] M. Yousefzadeh, S. A. Hosseini, and M. Farnaghi, “Spatiotemporally explicit earthquake prediction using deep neural network,”
Soil Dynamics and Earthquake Engineering, vol. 144, May 2021, doi: 10.1016/j.soildyn.2021.106663.
[44] R. Yuan, “An improved K-means clustering algorithm for global earthquake catalogs and earthquake magnitude prediction,”
Journal of Seismology, vol. 25, no. 3, pp. 1005–1020, Jun. 2021, doi: 10.1007/s10950-021-09999-8.
[45] S. Biswas, D. Kumar, and U. K. Bera, “Prediction of earthquake magnitude and seismic vulnerability mapping using artificial
intelligence techniques: a case study of Turkey,” Research Square Platform LLC, 2023. doi: 10.21203/rs.3.rs-2863887/v1.
[46] W. H. Al Yamani, M. Bisharah, H. H. Alumany, and N. A. Al Mohammadin, “Machine learning in seismic structural design: an
exploration of ANN and tabu-search optimization,” Asian Journal of Civil Engineering, vol. 25, no. 3, pp. 2367–2377, Apr. 2024,
doi: 10.1007/s42107-023-00913-w.
[47] Z. Bao, J. Zhao, P. Huang, S. Yong, and X. Wang, “A deep learning-based electromagnetic signal for earthquake magnitude
prediction,” Sensors, vol. 21, no. 13, Jun. 2021, doi: 10.3390/s21134434.
[48] D. Du and Z. Cui, “An evolutionary model for earthquake prediction considering time-series evolution and feature extraction and
its application,” Journal of Physics: Conference Series, vol. 2333, no. 1, Aug. 2022, doi: 10.1088/1742-6596/2333/1/012013.
[49] M. N. Islam, M. A. S. Khan, M. B. Munir, M. A. K. Azad, and T. Mustafy, “Seismic indicators-based earthquake magnitude
prediction for Bangladesh using machine learning algorithms,” in Proceedings of the 5rd International Conference on Civil
Structural and Transportation Engineering (ICCSTE’20), Nov. 2020, pp. 329-1-329–2. doi: 10.11159/iccste20.329.

BIOGRAPHIES OF AUTHORS

Sunarno specializes in Engineering, Instrumentation & Control, and

Engineering Physics. He is actively involved in the Engineering Physics, Nuclear
Engineering, and Instrumentation cluster within Science and Technology. His research
interests revolve around instrumentation for telemetry and tele-control, as well as early
warning systems for natural disasters. In terms of community services, he has conducted
several training programs. He can be contacted at email: [email protected].

Thomas Oka Pratama holds a Doctor of Environmental Science from

Universitas Gadjah Mada in 2021. He also received his S.T. and M.Eng. (Engineering
Physics) from Universitas Gadjah Mada in 2014 and 2021, respectively. He is currently a
lecturer in the Engineering Physics Program, in the Department of Nuclear Engineering and
Engineering Physics at Universitas Gadjah Mada, Indonesia. His research includes a
telemonitoring system and earthquake prediction based on the radon gas concentration and
groundwater level. He can be contacted at email: [email protected].

Faridah lecturer at the Department of Nuclear Engineering and Engineering

Physics, Universitas Gadjah Mada since 1999 with research interests in Instrumentation.
Faridah received her bachelor’s degree from the Department of Engineering Physics, Institut
Teknologi Sepuluh Nopember, and her master’s degree from the Microelectronics Program,
Nanyang Technological University, Singapore. Since 2015, Faridah has been active as a
researcher at INSGREEB and develops instrumentation in building systems. Faridah has
pursued a Doctoral’s degree in Environmental Science, at Gadjah Mada University, with a
research focus on the thermal comfort model of the living space environment for building
management systems by relying on telemonitoring systems in capturing data on
environmental parameters. She can be contacted at email: [email protected].

Earthquake magnitude prediction based on radon cloud data near Grindulu fault, Indonesia … (Sunarno)
836  ISSN: 1693-6930

Nugroho Ananto is an experienced professional in the fields of technology and

management. He earned a bachelor’s degree in mechanical engineering from Institut
Teknologi 10 Nopember Surabaya (ITS), a Master of Engineering in Mechanical Engineering
for Computer-Controlled Machinery from Osaka University, a Master of Management from
Institut Teknologi Bandung (ITB), and a Doctor of Business Management from Institut
Pertanian Bogor (IPB). Nugroho currently serves as the President Director/CEO of PT Energi
Primer Terbarukan (PRIME) since January 2021 and the Secretary General of the Indonesian
System Experts Association (PASINDO) since January 2019. Nugroho has made significant
contributions to instrumentation systems, renewable energy, strategic industries, and public
policy in Indonesia. He can be contacted at email: [email protected].

Hermin Kartika Sari researcher and lecturer at Politeknik Negeri Bandung

specializing in modelling, instrumentation systems, and measurement. She graduated with a
Bachelor’s degree (2019) and a Master’s in Engineering Physics (2021) from Universitas
Gadjah Mada. She is an expert in designing instrumentation systems for environments,
buildings, medical applications, and other fields requiring instrumentation systems. She also
holds several patents in these areas. She can be contacted at email:
[email protected].

Rony Wijaya is a professional engineer from CV Amakusa Instrumentation

Technology specializing in developing and innovating instrumentation systems. He obtained
his Bachelor of Engineering (2004) and Master of Engineering (2008) in industrial safety
engineering and Electrical Engineering at Universitas Gadjah Mada. In 2020, he earned the
Engineer (Ir) title through the professional engineering program at the Faculty of
Engineering, Universitas Gadjah Mada. In 2021, the Indonesian Engineers Association
awarded him the Intermediate Professional Engineer Certificate (IPM). His research areas
include the design of sensors and instrumentation, telemetry and tele-control systems, and
industrial safety. He also holds multiple patents related to disaster management and
transportation. He can be contacted at email: [email protected].

Memory Motivanisman Waruwu a professional engineer at Amakusa group

and a member of the Research Team at the Sensor and Tele-control System Laboratory,
Department of Nuclear Engineering and Engineering Physics, Universitas Gadjah Mada. He
completed his Bachelor of Engineering degree in Physics from the Faculty of Engineering,
Universitas Gadjah Mada in 2004 and his Master of Engineering degree in Industrial Safety
Engineering, Electrical Engineering from Universitas Gadjah Mada in 2014. In 2020, he
obtained the title of Engineer (Ir) from the professional engineering program at the Faculty
of Engineering, Universitas Gadjah Mada, and in 2021, he received the Intermediate
Professional Engineer Certificate (IPM) from the Indonesian Engineers Association. His
research focuses include Sensor and Instrumentation Design, Telemetry and Tele-control,
Medical Instrumentation, and Industrial Safety. He holds several patents in the fields of
medical instrumentation, disaster management, and transportation. He can be contacted at
email: [email protected].

TELKOMNIKA Telecommun Comput El Control, Vol. 23, No. 3, June 2025: 825-836