📄 medicine

Application of SARIMA and Hybrid SARIMA–ANN Models in Predicting and Forecasting Scorpionism Cases in Brazil

This study demonstrates that a hybrid SARIMA–Artificial Neural Network model outperforms traditional SARIMA approaches in forecasting monthly scorpionism cases in Brazil, revealing a persistent upward trend and seasonal patterns that underscore the urgent need for enhanced public health surveillance and preventive strategies.

Original authors: Amaury de Souza

Published 2026-06-30

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Amaury de Souza

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

The Big Picture: A Growing Problem

Imagine Brazil as a giant house where a specific type of guest—the scorpion—is throwing an increasingly wild party. Over the last few decades, the number of scorpion stings (called "scorpionism") has gone up significantly. This isn't just a bug problem; it's a public health emergency. The paper explains that this is happening because cities are growing fast, sanitation is sometimes lacking, and scorpions are getting really good at living in human neighborhoods.

The main goal of this study was to act like a weather forecaster, but instead of predicting rain, the researchers tried to predict when and how many people would get stung by scorpions in the future.

The Tools: Two Different Types of Crystal Balls

To make these predictions, the author used two different mathematical "crystal balls" (models) and compared them to see which one was better.

1. The SARIMA Model (The "Rigid Rule-Follower")
Think of the SARIMA model as a strict librarian who only follows a set schedule. It looks at the past data and says, "Okay, every year in July, the numbers go up, and in January, they go down. I will just follow that pattern."

How it works: It is great at spotting straight lines and repeating seasonal cycles (like the weather).
The paper's finding: This model worked well enough to see the general trend and the yearly cycles. It could tell us that scorpion stings usually spike during warmer, wetter months. However, it sometimes missed the sudden, crazy jumps in numbers because it assumes the world is very orderly.

2. The Hybrid SARIMA-ANN Model (The "Flexible Detective")
This model is a team-up. It takes the "Rule-Follower" (SARIMA) and pairs it with an "Artificial Neural Network" (ANN).

The Analogy: Imagine the SARIMA model does the easy math first. Then, it looks at the mistakes it made (the parts it couldn't explain) and hands those mistakes to the ANN. The ANN is like a detective trained to find hidden, messy patterns that don't follow rules. It looks at the "leftover" chaos and figures out the weird, non-linear reasons why the numbers spiked unexpectedly.
How it works: It combines the steady rhythm of the seasons with the ability to learn from complex, messy surprises.
The paper's finding: This team-up was the winner. It predicted the future more accurately than the Rule-Follower alone because it could handle both the predictable seasons and the unpredictable chaos.

What the Data Actually Showed

The researchers looked at data from 2007 to 2025. Here is what they found:

The Trend is Up: The number of scorpion stings has been climbing steadily, like a staircase going up. It's not just a temporary spike; it's a long-term growth.
The Seasonal Rhythm: There is a clear beat to the problem. Just like flowers bloom in spring, scorpion stings happen more often in warmer, wetter months. The scorpions get more active, and people get stung more.
The Future Prediction: When they used their models to look ahead (forecasting into 2026), both models agreed on one thing: The party isn't stopping. The number of stings is expected to stay high. The "Flexible Detective" (Hybrid model) gave a slightly smoother, more reliable prediction, but both warned that Brazil needs to be ready for a lot of stings in the coming year.

Why This Matters (According to the Paper)

The paper argues that we can't just guess when the next wave of stings will happen. We need a plan.

Early Warning: If we know the "season" is coming, hospitals can prepare more medicine (antivenom) and doctors can get ready.
Prevention: The study suggests that fixing the root causes—like better trash management, cleaner sewers, and educating people on how to avoid scorpions—is the only way to stop the numbers from rising.

The Limitations (What the Paper Admitted)

The author was honest about the flaws in their crystal ball:

Missing Data: Sometimes, people in remote areas don't report stings, so the numbers might be lower than reality.
No Weather Variables: The models looked at the history of stings, but they didn't explicitly plug in daily temperature or rain data. The paper suggests that future models should include these specific weather details to be even smarter.

The Bottom Line

This paper is a proof-of-concept that combining old-school statistics with modern AI creates a better tool for predicting scorpion attacks in Brazil. The "Hybrid" model is the best tool currently available to help the country prepare for a future where scorpion stings remain a major health challenge.

Technical Summary: Application of SARIMA and Hybrid SARIMA–ANN Models in Predicting and Forecasting Scorpionism Cases in Brazil

Problem Statement
Scorpionism has emerged as a critical environmental and public health challenge in Brazil, characterized by a substantial increase in reported cases over recent decades. This rise is driven by urban expansion, inadequate sanitation, climate variability, and the ecological adaptation of Tityus species (particularly T. serrulatus, T. stigmurus, and T. bahiensis) to anthropogenic environments. While the epidemiological burden is high, there is a scarcity of studies applying advanced time-series and machine learning approaches to predict the temporal dynamics of scorpion envenomation. Traditional linear models often fail to capture the complex nonlinear interactions inherent in epidemiological data, necessitating more robust forecasting frameworks to support public health surveillance, resource allocation, and preventive strategies.

Methodology
This retrospective quantitative study analyzed monthly aggregated scorpionism cases in Brazil from 2007 to 2025. The research employed a two-stage modeling approach:

SARIMA Modeling: A Seasonal Autoregressive Integrated Moving Average (SARIMA) model was first fitted to the time series following the Box–Jenkins methodology. The optimal model, SARIMA(1,1,1)(0,1,1)₁₂, was selected based on the Akaike Information Criterion (AIC) and Bayesian Information Criterion (BIC). The model utilized a seasonal period ( $s$ ) of 12 months to account for annual cycles.
Hybrid SARIMA–ANN Modeling: To address potential nonlinear components not captured by the linear SARIMA structure, the residuals from the optimal SARIMA model were extracted and modeled using an Artificial Neural Network (ANN). Specifically, a Multilayer Perceptron (MLP) with a logistic activation function was trained on these residuals. The final hybrid forecast was constructed by summing the linear SARIMA predictions and the nonlinear ANN residual predictions.
Evaluation and Validation: Model performance was assessed using Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), and Mean Absolute Percentage Error (MAPE). The adequacy of the SARIMA model was verified via the Ljung–Box test for white noise residuals. The predictive superiority of the hybrid model over the single SARIMA model was statistically validated using the Diebold–Mariano (DM) test.

Key Contributions

Comparative Modeling: The study provides a direct comparison between a standard linear time-series approach (SARIMA) and a hybrid machine learning approach (SARIMA–ANN) specifically for scorpionism data in Brazil, a domain where such predictive modeling is currently limited.
Hybrid Framework Implementation: It demonstrates the technical feasibility of decomposing epidemiological time series into linear and nonlinear components, modeling them separately, and recombining them to enhance forecasting accuracy.
Forecasting Horizon: The study extends beyond descriptive analysis to generate point forecasts for the period of 2026, offering a quantitative basis for anticipating future incidence trends.

Results

Temporal Trends: Exploratory analysis confirmed a marked increasing trend in scorpionism cases from 2007 to 2025, alongside pronounced seasonality with peaks occurring during warmer and wetter months.
Model Performance: The SARIMA(1,1,1)(0,1,1)₁₂ model successfully captured the general trend and seasonal oscillations. However, residual diagnostics indicated the presence of remaining nonlinear structures.
Superiority of Hybrid Model: The hybrid SARIMA–ANN model demonstrated statistically superior predictive performance compared to the standalone SARIMA model, as evidenced by lower RMSE, MAE, and MAPE values and a significant result in the Diebold–Mariano test. The hybrid model showed improved capability in reproducing seasonal peaks and abrupt incidence fluctuations.
Future Projections: Both models forecasted the persistence of elevated scorpionism incidence levels throughout 2026, maintaining the historical seasonal pattern. The hybrid model produced smoother and more stable forecasts while preserving the seasonal structure.

Significance and Claims
The paper asserts that the hybrid SARIMA–ANN approach constitutes a promising tool for epidemiological forecasting of scorpionism in Brazil. By integrating classical statistical methods with artificial intelligence, the study claims to offer a more accurate and reliable method for capturing the complex linear and nonlinear dynamics of scorpion envenomation.

The authors position these findings as a contribution to public health planning, suggesting that such models can support the implementation of early-warning systems. Specifically, the forecasts can assist health authorities in anticipating high-risk periods, optimizing hospital preparedness, allocating antivenom stocks, and organizing targeted preventive campaigns. The study concludes that while the models show satisfactory predictive capability, the growing burden of scorpionism necessitates strengthened surveillance, environmental management, and preventive strategies, with the hybrid model serving as a valuable asset for decision-making processes.

Limitations Noted
The authors acknowledge that the study relies on aggregated secondary data, which may be subject to underreporting and notification delays. Furthermore, the models did not explicitly incorporate climatic, environmental, or socioeconomic variables (e.g., temperature, precipitation, population density), which could potentially enhance explanatory power in future frameworks. Consequently, long-term forecasts are advised to be interpreted with caution.