Time Series Datasets Research Articles

Transport-Related Synthetic Time Series: Developing and Applying a Quality Assessment Framework

Data scarcity and privacy concerns in various fields, including transportation, have fueled a growing interest in synthetic data generation. Synthetic datasets offer a practical solution to address data limitations, such as the underrepresentation of minority classes, while maintaining privacy when needed. Notably, recent studies have highlighted the potential of combining real and synthetic data to enhance the accuracy of demand predictions for shared transport services, thereby improving service quality and advancing sustainable transportation. This study introduces a systematic methodology for evaluating the quality of synthetic transport-related time series datasets. The framework incorporates multiple performance indicators addressing six aspects of quality: fidelity, distribution matching, diversity, coverage, and novelty. By combining distributional measures like Hellinger distance with time-series-specific metrics such as dynamic time warping and cosine similarity, the methodology ensures a comprehensive assessment. A clustering-based evaluation is also included to analyze the representation of distinct sub-groups within the data. The methodology was applied to two datasets: passenger counts on an intercity bus route and vehicle speeds along an urban road. While the synthetic speed dataset adequately captured the diversity and patterns of the real data, the passenger count dataset failed to represent key cluster-specific variations. These findings demonstrate the proposed methodology’s ability to identify both satisfactory and unsatisfactory synthetic datasets. Moreover, its sequential design enables the detection of gaps in deeper layers of similarity, going beyond basic distributional alignment. This work underscores the value of tailored evaluation frameworks for synthetic time series, advancing their utility in transportation research and practice.

Predicting the impact of climate change on building energy consumption by using data-driven approaches

Buildings are complex thermodynamic entities that account for a large proportion of energy consumption. This work explores the application of data-driven models in order to forecast the building energy consumption over long time horizons. Long Short-Term Memory and Random Forest are used to forecast hourly heating and cooling energy consumption in the Urban Sciences Building, which is located in the city of Newcastle upon Tyne, UK. A synthetic time-series dataset is constructed using (a) a validated EnergyPlus model and (b) operational data hosted by Newcastle Urban Observatory. The energy consumption forecast is made using future climate data that represent a high emission future scenario during a typical year, that is 2030, and 2080. The experimental results suggest that, on average over the next six decades, there will be a 45% reduction in annual heating load, accompanied by a 680% increase in annual cooling load. These early findings indicate a notable shift in building thermal performance, a rise in overall building energy demand under the more drastic future weather scenarios, and a sharper emphasis on designing building envelopes and energy systems to include decarbonised cooling solutions.

Forty-Year Time Series of Anthropogenic Uranium in Fucus vesiculosus from Kattegat: Validating 236U Releases from European Nuclear Reprocessing Plants.

An accurate input function of uranium-236 (236U) is essential for its numerous applications in environmental and ocean studies. This work reveals potential overestimation on 236U releases from the Sellafield (SF) nuclear reprocessing plant (RP) by earlier estimations using shells. We report a 40-year time series dataset of uranium isotopes in seaweed seasonally collected in the Kattegat, downstream of radioactive discharges from the European RPs, namely La Hague (LH) and SF. Comparison between our measured 236U concentrations in seaweed and model-simulated values derived from 236U releases of LH and SF suggests that the previously reconstructed SF discharges may have nearly 1 order of magnitude overestimation. Such overestimation would introduce large uncertainties when 236U is used as a point source in environmental tracer applications and ocean model validation. We foresee that this work will significantly improve the applications of radioisotope tracers in oceanic studies such as the spreading of marine pollutants, ocean mixing, and circulation, especially in the regions of the North Atlantic and Arctic Oceans.

Evaluation-Free Time-Series Forecasting Model Selection via Meta-Learning

Time-series forecasting models are invariably used in a variety of domains for crucial decision-making. Traditionally these models are constructed by experts with considerable manual effort. Unfortunately, this approach has poor scalability while generating accurate forecasts for new datasets belonging to diverse applications. Without access to skilled domain-knowledge, one approach is to train all the models on the new time-series data and then select the best one. However, this approach is nonviable in practice. In this work, we develop techniques for fast automatic selection of the best forecasting model for a new unseen time-series dataset, without having to first train (or evaluate) all the models on the new time-series data to select the best one. In particular, we develop a forecasting meta-learning approach called AutoForecast that allows for the quick inference of the best time-series forecasting model for an unseen dataset. Our approach learns both forecasting models performances over time horizon of the same dataset and task similarity across different datasets. The experiments demonstrate the effectiveness of the approach over state-of-the-art (SOTA) single and ensemble methods and several SOTA meta-learners (adapted to our problem) in terms of selecting better forecasting models (i.e., 2X gain) for unseen tasks for univariate and multivariate testbeds. AutoForecast has also significant reduction in inference time compared to the naïve approach (doing inference using all possible models and then selecting the best one), with median of 42X across the two testbeds. We release our meta-learning database corpus (348 datasets), performances of the 322 forecasting models on the database corpus, meta-features, and source codes for the community to access them for forecasting model selection and to build on them with new datasets and models which can help advance automating time-series forecasting problem. In our released database corpus, we unveil new traces of Adobe computing cluster usage for production workloads.

A novel multivariate time series dataset of outdoor sport activities

This study introduces a novel multivariate time series dataset of 228 outdoor sport activities recorded by individual non-competitive athlete in uncontrolled environments. The dataset includes three features: Heart Rate, Speed, and Altitude, and covers five sport categories: walking, running, skiing, roller-skiing, and biking. The data was collected using two types of Garmin sport watches. The original dataset was carefully pre-processed using typical data cleansing methods such as gaps filling, and value format transformations. Furthermore, activity filtering was implemented for missing sensor value data and using domain knowledge of sport categories. Full length sequences, varying from 10 min to several hours, were split into equal length segments, approximately 1 min. To address the small number of instances data was augmented using several consecutive segments from the same activity. However, only a small part of the whole original data was used as a computational cost–information gain tradeoff. Three-dimensional dataset is divided into three parts, each dimension to its own comma separated value (CSV) file. The dataset aims to provide a unique resource for researchers and practitioners in the field of sports science, human performance analysis, and activity recognition. It aims to complement the very limited or non-existent publicly available sport activity datasets.

A Topological Approach to Enhancing Consistency in Machine Learning via Recurrent Neural Networks

The analysis of continuous events for any application involves the discretization of an event into sequences with potential historical dependencies. These sequences represent time stamps or samplings of a continuous process collectively forming a time series dataset utilized for training recurrent neural networks (RNNs) such as Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU) for pattern prediction. The challenge is to ensure that the estimates from the trained models are consistent in the same input domain for different discretizations of the same or similar continuous history-dependent events. In other words, if different time stamps are used during the prediction phase after training, the model is still expected to give consistent predictions based on the knowledge it has learned. To address this, we present a novel RNN transition formula intended to produce consistent estimates in a wide range of engineering applications. The approach was validated with synthetically generated datasets in 1D, 2D, and 3D spaces, intentionally designed to exhibit high non-linearity and complexity. Furthermore, we have verified our results with real-world datasets to ensure practical applicability and robustness. These assessments show the ability of the proposed method, which involves restructuring the mathematical structure and extending conventional RNN architectures, to provide reliable and consistent estimates for complex time series data.

Bidirectional f-Divergence-Based Deep Generative Method for Imputing Missing Values in Time-Series Data

Imputing missing values in high-dimensional time-series data remains a significant challenge in statistics and machine learning. Although various methods have been proposed in recent years, many struggle with limitations and reduced accuracy, particularly when the missing rate is high. In this work, we present a novel f-divergence-based bidirectional generative adversarial imputation network, tf-BiGAIN, designed to address these challenges in time-series data imputation. Unlike traditional imputation methods, tf-BiGAIN employs a generative model to synthesize missing values without relying on distributional assumptions. The imputation process is achieved by training two neural networks, implemented using bidirectional modified gated recurrent units, with f-divergence serving as the objective function to guide optimization. Compared to existing deep learning-based methods, tf-BiGAIN introduces two key innovations. First, the use of f-divergence provides a flexible and adaptable framework for optimizing the model across diverse imputation tasks, enhancing its versatility. Second, the use of bidirectional gated recurrent units allows the model to leverage both forward and backward temporal information. This bidirectional approach enables the model to effectively capture dependencies from both past and future observations, enhancing its imputation accuracy and robustness. We applied tf-BiGAIN to analyze two real-world time-series datasets, demonstrating its superior performance in imputing missing values and outperforming existing methods in terms of accuracy and robustness.

A data-driven approach to study temporal characteristics of COVID-19 infection and death Time Series for twelve countries across six continents

BackgroundIn this work, we implement a data-driven approach using an aggregation of several analytical methods to study the characteristics of COVID-19 daily infection and death time series and identify correlations and characteristic trends that can be corroborated to the time evolution of this disease. The datasets cover twelve distinct countries across six continents, from January 22, 2020 till March 1, 2022. This time span is partitioned into three windows: (1) pre-vaccine, (2) post-vaccine and pre-omicron (BA.1 variant), and (3) post-vaccine including post-omicron variant. This study enables deriving insights into intriguing questions related to the science of system dynamics pertaining to COVID-19 evolution.MethodsWe implement a set of several distinct analytical methods for: (a) statistical studies to estimate the skewness and kurtosis of the data distributions; (b) analyzing the stationarity properties of these time series using the Augmented Dickey-Fuller (ADF) tests; (c) examining co-integration properties for the non-stationary time series using the Phillips-Ouliaris (PO) tests; (d) calculating the Hurst exponent using the rescaled-range (R/S) analysis, along with the Detrended Fluctuation Analysis (DFA), for self-affinity studies of the evolving dynamical datasets.ResultsWe notably observe a significant asymmetry of distributions shows from skewness and the presence of heavy tails is noted from kurtosis. The daily infection and death data are, by and large, nonstationary, while their corresponding log return values render stationarity. The self-affinity studies through the Hurst exponents and DFA exhibit intriguing local changes over time. These changes can be attributed to the underlying dynamics of state transitions, especially from a random state to either mean-reversion or long-range memory/persistence states.ConclusionsWe conduct systematic studies covering a widely diverse time series datasets of the daily infections and deaths during the evolution of the COVID-19 pandemic. We demonstrate the merit of a multiple analytics frameworks through systematically laying down a methodological structure for analyses and quantitatively examining the evolution of the daily COVID-19 infection and death cases. This methodology builds a capability for tracking dynamically evolving states pertaining to critical problems.

Calibrated Ecosystem Models Cannot Predict the Consequences of Conservation Management Decisions.

Ecosystem models are often used to predict the consequences of management interventions in applied ecology and conservation. These models are often high-dimensional and nonlinear, yet limited data are available to calibrate or validate them. Consequently, their utility as decision-support tools is unclear. In this paper, we calibrate ecosystem models to time series data from 110 different experimental microcosm ecosystems, each containing three to five interacting species. Then, we assess their ability to predict the consequences of management interventions. Our results show that for each time series dataset, multiple divergent parameter sets offer equivalent, good fits. However, these models have poor predictive accuracy when forecasting future dynamics or when predicting how the ecosystem will respond to management intervention. Closer inspection reveals that the models fail because calibration cannot determine the nature of the interspecific interactions. Our findings question whether ecosystem models can support applied ecological decision-making when calibrated against real-world datasets.

A Survey of Time Series Data Forecasting Methods Based on Deep Learning

Time Series Forecasting (TSF) involves predicting future values and trends of data at specific points or periods by analyzing historical patterns, such as trends and seasonality. With the advent of IoT sensors, traditional machine learning approaches struggle to handle massive time series datasets. Recently, deep learning algorithms, exemplified by convolutional neural networks (CNNs), recurrent neural networks (RNNs), and Transformer models, have made significant progress in time series forecasting tasks. This paper reviews the common features of time series data, relevant datasets, and evaluation metrics for models. It also conducts experimental comparisons of various forecasting algorithms, focusing on time and algorithmic architectures. This paper conducts prediction experiments on several deep learning models using the ETT dataset and presents the final results. We evaluate model performance using metrics like Mean Absolute Error (MAE) and Mean Squared Error (MSE). We highlight the strengths and weaknesses of deep learning-based TSF methods. Major deep learning-based time series forecasting methods are introduced and compared. Finally, challenges and future research directions in applying deep learning to time series forecasting are discussed.

An Analysis And Forecasting The Foodstuffs Prices In Surabaya Traditional Market Using LSTM

Food is one of the essential things in society. Foodstuffs prices are important factors in the stability economy. In Indonesian society, some foodstuffs, e.g., rice, beef, chicken egg, cooking oil, and sugar are the main ingredients in their cuisine. Analyzing and predicting the foodstuffs price is interesting job. This research is conducted to develop models for forecasting the price of rice, beef, chicken egg, cooking oil, and sugar. It implements the Long Short-Term Memory (LSTM) model and a daily time-series dataset from a traditional market in Surabaya. Surabaya is the capital city of East Java province, and it is one of the densest cities in Indonesia. The experiments run univariate time-series forecasting. The experimental results show that LSTM works well to forecast the price of rice, beef, chicken egg, cooking oil, and sugar. The evaluation results obtain MAPE scores as 0.12%, 0.03%, 0.72%, 0.36%, and 0.08% for models of rice, beef, chicken egg, cooking oil, and sugar, respectively. The annual average price of beef, chicken egg, and cooking oil show an increasing trend and those foodstuffs have positive correlations with each other.

A comprehensive time-series dataset linked to cyanobacterial blooms in Lake Taihu

Lake Taihu has a history of recurrent harmful cyanobacterial blooms. There is a need to better understand the aquatic ecosystem of Lake Taihu in order to improve methods for controlling the cyanobacterial blooms. Based on the field measurement and satellite remote sensing, we produced and collected a time-series dataset, including the water quality, bio-optics, climate, and anthropogenic data of Lake Taihu (THQBCA), which could provide comprehensive information regarding cyanobacterial blooms. The THQBCA dataset contains 26 variables organized into four categories: water quality, bio-optics, climate, and anthropogenic data. The water quality and climate data are field measured data with sampling frequency from daily to quarterly, and bio-optics and anthropogenic data are satellite-derived annual data. The dataset spans more than 15 years (8 of which cover approximately 35 years, 4 of which cover 20 years), and the spatial resolutions of the satellite-derived data range from 30 m to 500 m. This dataset is expected to advance research on evaluating and predicting cyanobacterial blooms, and support science-based management decisions for sustainable ecological development.

Time-Series Representation Feature Refinement with a Learnable Masking Augmentation Framework in Contrastive Learning.

In this study, we propose a novel framework for time-series representation learning that integrates a learnable masking-augmentation strategy into a contrastive learning framework. Time-series data pose challenges due to their temporal dependencies and feature-extraction complexities. To address these challenges, we introduce a masking-based reconstruction approach within a contrastive learning context, aiming to enhance the model's ability to learn discriminative temporal features. Our method leverages self-supervised learning to effectively capture both global and local patterns by strategically masking segments of the time-series data and reconstructing them, which aids in revealing nuanced temporal dependencies. We utilize learnable masking as a dynamic augmentation technique, which enables the model to optimize contextual relationships in the data and extract meaningful representations that are both context-aware and robust. Extensive experiments were conducted on multiple time-series datasets, including SleepEDF-78, 20, UCI-HAR, achieving improvements of 2%, 2.55%, and 3.89% each and similar performance on Epilepsy in accuracy over baseline methods. Our results show significant performance gains compared to existing methods, highlighting the potential of our framework to advance the field of time-series analysis by improving the quality of learned representations and enhancing downstream task performance.

Facial Biosignals Time–Series Dataset (FBioT): A Visual–Temporal Facial Expression Recognition (VT-FER) Approach

Visual biosignals can be used to analyze human behavioral activities and serve as a primary resource for Facial Expression Recognition (FER). FER computational systems face significant challenges, arising from both spatial and temporal effects. Spatial challenges include deformations or occlusions of facial geometry, while temporal challenges involve discontinuities in motion observation due to high variability in poses and dynamic conditions such as rotation and translation. To enhance the analytical precision and validation reliability of FER systems, several datasets have been proposed. However, most of these datasets focus primarily on spatial characteristics, rely on static images, or consist of short videos captured in highly controlled environments. These constraints significantly reduce the applicability of such systems in real-world scenarios. This paper proposes the Facial Biosignals Time–Series Dataset (FBioT), a novel dataset providing temporal descriptors and features extracted from common videos recorded in uncontrolled environments. To automate dataset construction, we propose Visual–Temporal Facial Expression Recognition (VT-FER), a method that stabilizes temporal effects using normalized measurements based on the principles of the Facial Action Coding System (FACS) and generates signature patterns of expression movements for correlation with real-world temporal events. To demonstrate feasibility, we applied the method to create a pilot version of the FBioT dataset. This pilot resulted in approximately 10,000 s of public videos captured under real-world facial motion conditions, from which we extracted 22 direct and virtual metrics representing facial muscle deformations. During this process, we preliminarily labeled and qualified 3046 temporal events representing two emotion classes. As a proof of concept, these emotion classes were used as input for training neural networks, with results summarized in this paper and available in an open-source online repository.

Socio-Demographic Determinants of Mortality in Bangladesh: A Time Series Analysis

Understanding the factors influencing the death rate in Bangladesh is essential for effective policy-making and addressing demographic challenges. This study aims to examine both the short-term dynamics and long-term equilibrium relationships between socio-demographic factors and the death rate of Bangladesh over a given period. It focuses on four key factors: adolescence fertility rate (AFR), life expectancy rate at birth, birth rate, and infant mortality rate (IMR). Taking an annual time series dataset from 1975 to 2021 from the World Bank (WB), the study uses the Johansen co-integration test, vector error correction model (VECM), and Granger causality test to analyze the relationships between these factors and death rate. The findings reveal short-term and long-term equilibrium links between the socio-demographic parameters and death rate. A substantial negative correlation exists between death rate and AFR and life expectancy. Additionally, a positive correlation is observed between the death rate and birth rate, and IMR. These findings provide valuable insights for decision-makers to address the death rate in Bangladesh, bridging a research gap and guiding effective socio-demographic policies. International Journal of Statistical Sciences, Vol. 24(2), November, 2024, pp 35-54

Air pollution forecasting: Application of machine learning models to estimate PM2.5 index

In the context of ongoing industrialization, air pollution has become an urgent global problem, particularly severe in large cities such as Hanoi (Vietnam), Beijing (China), and others. Air pollution, especially the concentration of fine particulate matter (PM2.5), is not only harmful to human health but also has significant negative impacts on the environment, economy, and quality of life. This study aims to enhance the ability to predict air pollution levels more accurately. By using machine learning models, meteorologists can better predict air pollution levels and propose more effective mitigation solutions. The article utilizes a multivariate time series dataset, including meteorological and air pollution indices from Beijing, China, from 2010 to 2014. Machine learning models such as Lasso Regression, Support Vector Regression, Random Forest, XGBoost, and, notably, a Stack Model combining the four aforementioned models, are evaluated. The performance of these models is measured using statistical indicators such as Root Mean Square Error (RMSE), Mean Absolute Error (MAE), and Coefficient of Determination (R²). Among these models, the Stack model provides the most accurate predictions for the PM2.5 index.

Machine Learning-Based Summer Crops Mapping Using Sentinel-1 and Sentinel-2 Images

Accurate crop type mapping using satellite imagery is crucial for food security, yet accurately distinguishing between crops with similar spectral signatures is challenging. This study assessed the performance of Sentinel-2 (S2) time series (spectral bands and vegetation indices), Sentinel-1 (S1) time series (backscattering coefficients and polarimetric parameters), alongside phenological features derived from both S1 and S2 time series (harmonic coefficients and median features), for classifying sunflower, soybean, and maize. Random Forest (RF), Multi-Layer Perceptron (MLP), and XGBoost classifiers were applied across various dataset configurations and train-test splits over two study sites and years in France. Additionally, the InceptionTime classifier, specifically designed for time series data, was tested exclusively with time series datasets to compare its performance against the three general machine learning algorithms (RF, XGBoost, and MLP). The results showed that XGBoost outperformed RF and MLP in classifying the three crops. The optimal dataset for mapping all three crops combined S1 backscattering coefficients with S2 vegetation indices, with comparable results between phenological features and time series data (mean F1 scores of 89.9% for sunflower, 76.6% for soybean, and 91.1% for maize). However, when using individual satellite sensors, S1 phenological features and time series outperformed S2 for sunflower, while S2 was superior for soybean and maize. Both phenological features and time series data produced close mean F1 scores across spatial, temporal, and spatiotemporal transfer scenarios, though median features dataset was the best choice for spatiotemporal transfer. Polarimetric S1 data did not yield effective results. The InceptionTime classifier further improved classification accuracy over XGBoost for all crops, with the degree of improvement varying by crop and dataset (the highest mean F1 scores of 90.6% for sunflower, 86.0% for soybean, and 93.5% for maize).

Identification of optimal Sentinel-1 SAR polarimetric parameters for forest monitoring in Czechia

Time series analysis of synthetic aperture radar data (SAR) offers a systematic, dynamic and comprehensive way to monitor forests. The main emphasis of this study is on the identification of the most suitable and best performing Sentinel-1 SAR polarimetric parameters for forest monitoring. This is accomplished through: 1) a pairwise correlation analysis of SAR polarimetric parameters, multispectral optical vegetation indices and ancillary data, 2) a univariate binary time series classification for differentiation between forest types and 3) a visual exploration of time series. For this purpose, 600 validated broad-leaved and 600 coniferous forest areas in Czechia were used. Nine different SAR polarimetric parameters were examined, including VH and VV polarizations, VV/VH and VH/VV polarization ratios, the Radar Vegetation Index, Radar Forest Degradation Index, polarimetric radar vegetation index and the original and modified versions of the dual polarimetric SAR vegetation index. The pairwise correlation analysis revealed that most of the derived SAR polarimetric parameters were functions of each other with nearly identical behavior (r > |0.96|). The strongest correlation of r ~0.50 between SAR and optical features was found for broad-leaved forest for VV/VH and VH/VV. The highest overall accuracy in the time series classification of forest types was achieved by VH (76%), while for VV, VV/VH and VH/VV it was higher than 60%. Furthermore, the time series analysis of these parameters showed seasonal behaviors of the SAR features in both forest types. These results demonstrated the high relevance of using VH, VV, VV/VH and VH/VV time series in forest monitoring compared to other SAR polarimetric parameters. This study also introduces a novel pipeline to generate multi-modal time series datasets in Google Earth Engine (MMTS-GEE), used to generate data for the analysis. MMTS-GEE combines spatially and temporally aligned SAR and multispectral data, extended with topographic and weather data, and a land cover class label. Its high versatility enables its use in time series analyses, intercomparisons and in machine learning applications for tabular time series data. The GEE code for the proposed tool and analysis is freely available to the research community.

More pubs-More problems? A longitudinal analysis of Swedish panel data.

The aim of the study is to estimate the association between bar density and nighttime emergency calls to the police. We used a pooled cross-sectional time-series data set covering the Swedish 290 municipalities spanning the time period 2012-2021. As outcome we used nighttime emergency calls to the police and daytime emergency calls to the police as control variable. Bar density was measured as number of serving establishments licensed to serve alcohol after 1 am. The municipalities were classified into three socio-economic (SES) categories: low, middle and high-SES areas. We used survey data to estimate the prevalence of heavy episodic drinking (HED) for each of the three SES areas. We applied fixed-effects modelling to estimate the association between bar density and nighttime police calls, implying that only within-unit variation was exploited. Findings suggested that a 10% increase in bar density would result in a 0.46% (p = 0.001) increase in nighttime police calls. There was a marked gradient in the bar effect across SES areas; in high-SES areas the effect was not significant, while the effect was twice as strong in the low-SES areas as in mid-SES areas. We also found that the lower the status of the SES area, the higher the prevalence of HED. We found a significantly positive association between bar density and nighttime police calls. However, the association was markedly stronger in low-SES areas (which were also characterised by an elevated HED prevalence) than in high-SES areas (distinguished by a lower HED prevalence).

Modeling Techniques for Estimating Soil Radon Concentrations

Abstract This paper describes the comprehensive analysis of soil radon concentrations through the application of the Auto Regressive Integrated Moving Average (ARIMA) Model. The study utilizes information on radon gathered from March 2013 to May 2016, employing LR-115 type II detectors. Radon and thoron, prominent constituents of natural radioactivity in the Earth’s crust, are pivotal subjects in geophysics and indoor measurements. The investigation focuses on deriving the predicted values of radon concentrations, employing various ARIMA models. Specifically, four distinct models: ARIMA (3, 0, 7), ARIMA (4, 0, 7), ARIMA (3, 0, 6), and ARIMA (4, 0, 6) are constructed. Evaluation of the auto-correlation function (ACF) and partial auto correlation function (PACF) plots suggests that ARIMA (3, 0, 7) and ARIMA (4, 0, 7) stand out as promising candidates for effectively capturing the data dynamics. To validate the efficacy of each model and ascertain the most suitable one, a comparative analysis between the estimated and real values is conducted. This involves the implementation of a linear fit test and an F-test. The statistical results decisively indicate that ARIMA (3, 0, 7) outperforms ARIMA (4, 0, 7), establishing its superiority as a better model to forecast radon in soil concentrations in the given time series data. This comprehensive approach demonstrates the significance of precise predictive models in spotting deviations in time series datasets and advances the understanding of soil radon variation.