IRAQI JOURNAL OF STATISTICAL SCIENCES

A Comparative Analysis of ARIMA and LSTM Models for Forecasting Non-Stationary Financial Time Series

2026-02-09T17:36:12+00:00

Stock prices have become relatively wilder and nonlinear in nature over time. As such, outcome forecasting ability is essential for decision-making in finance. This research compares an adaptive Long Short-Term Memory neural net to the traditional ARIMA model in terms of predictability for a commonly used financial time series that is known to be non-stationary. Preprocessing using Normalizer and Augmented Dickey-Fuller (ADF) test for stationarity was carried out on Historical daily stock Close Price data collected from Yahoo Finance. Based on ACF/PACF analysis, an ARIMA (5,1,0) model was developed, while a multi-layer LSTM captured long-run dependencies. The Mean Squared Error (MSE), Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), and R-square (R²) were used to evaluate the models. The results revealed that LSTM outperformed ARIMA, with MSEs of 0.0023 and 0.0456, respectively. In addition, the LSTM model was more robust against sudden price variations, with an R² value of 0.92 versus 0.857 for ARIMA. Such findings show that while ARIMA remains useful for detecting linear trends, adaptive deep learning models indicate that LSTM is far more effective in the case of dynamic, non-stationary environments. Future studies must thus explore hybrid architectures that take advantage of both approaches.

Efficient XML Parsing: Enhancing Efficiency Through Design and Analysis

2026-02-09T17:44:01+00:00

The development and analysis of UML class diagrams are fundamental aspects of software engineering, providing insights into system structure and design. This paper introduces a novel XML parser designed to efficiently parse and classify UML class diagrams, leveraging XML’s structured format for improved data extraction and evaluation. The proposed parser addresses limitations identified in previous research, particularly in handling large and complex UML structures, performance optimization, and integration with machine learning models for advanced diagram classification. The parser’s ability to process detailed relationships and hierarchies within the diagrams enhances the accuracy of classification, and the integration with machine learning models facilitates automated analysis and prediction of diagram quality. The results of this parser are presented as inputs for further machine learning models, contributing to enhanced software development processes. Through systematic testing and comparison with existing methods, this paper demonstrates the parser’s superior efficiency and scalability, making it a valuable tool for both UML diagram analysis and future research in software engineering.

Improved Pareto Set Based Method for Solving -Objective Linear Programming Problems

2026-02-09T17:51:56+00:00

A basic idea in mathematics, convex combinations are especially important in linear algebra, convex analysis, and optimization. They offer a method for creating new points from a given set while maintaining convexity, which is a crucial characteristic in many applied and mathematical domains. In this paper, we improved and generalized an idea for solving multi-objective function to find a compromise solution. The method focused on using convex combination of some points namely, positive efficient points which is a new definition. The results providing acceptable solutions for decision-makers. Comparing it with other methods, the method gives range of solutions as well as, can be used in the case of the individual optimal solutions are on distinct extreme points.

Identifying the Most Important Symptoms Indicative of Diagnosing Suspected Cases of Coronavirus using a Hesitant Fuzzy Sample

2026-02-09T17:58:42+00:00

In this article, the traditional method and the fuzzy sampling method were used to identify the most important symptoms indicative of suspected infection with the Coronavirus using a questionnaire prepared for the purpose of collecting information about the phenomenon studied. The questionnaire was distributed in three hospitals affiliated with the Babylon Health Department to doctors with specific specialties (chest - respiratory - internal medicine), with (20) questionnaires in Marjan Teaching Hospital, (20) questionnaires in Imam Al-Sadiq (peace be upon him) Teaching Hospital, and (15) questionnaires in Al-Hashimiya General Hospital. Thus, the total number of individuals in the research sample was (55) specialist doctors. It was found that the fuzzy sampling method was more accurate than the traditional method in identifying the most important symptoms indicative of suspected infection with the Coronavirus, which were distinguished as main signs of infection, namely fever, shortness of breath, loss of the sense of smell or taste, and muscle pain. As for the symptoms of nasal congestion, runny nose, nausea or vomiting, and diarrhea, they are the least common in diagnosis. The symptoms of dry cough, fatigue, headache, and sore throat are not clearly influential in determining infection.

Variables Selection in Inverse Gaussian Regression Model using Modified Heuristic Search Algorithm

2026-03-17T11:46:29+00:00

The inverse Gaussian regression model is one of the most widely recognized models, frequently used across various applications. It is part of the generalized linear model families and serves as a foundational model. Like other regression models, it may include numerous independent variables, which can negatively impact both the model’s accuracy and the simplicity of interpreting its results. This study aims to apply the modified invasive weed optimization algorithm and compare it with other methods for variable selection in the inverse Gaussian regression model, using both simulations and real-world data. The Monte Carlo simulation approach was employed, setting sample sizes n to four different values—30, 50, 100, and 150—to facilitate comparisons across sample sizes (small, medium, large). Results indicated that the proposed method reduces the mean square error of the model and outperforms previously used techniques. The AIC method emerged as the least effective in variable selection, as it yielded the highest prediction error (PE) and tended to select irrelevant explanatory variables.

Decoding Labor Market Efficiency in Arab Economies:A Statistical Benchmarking of Egypt’s Performance

2026-03-17T12:51:12+00:00

This paper presents a comprehensive statistical examination of labor market efficiency across Arab economies, using Egypt as a benchmark case to assess regional disparities in competitiveness. The study integrates historical data from the World Economic Forum’s Global Competitiveness Index (GCI 2017-2018) with the most recent insights from the IMD World Competitiveness Yearbook 2024 (WCY 2024), providing a multidimensional perspective on how labor-market structures, demographic dynamics, and institutional frameworks jointly shape competitive performance. Through comparative quantitative analysis, the research explores ten GCI labor efficiency indicators including wage flexibility, professional management reliance, female participation, and talent retention contrasted against IMD 2024 factor-level rankings in Business Efficiency and labor-market adaptability for selected Arab economies. Findings reveal a persistent performance gap between Egypt and the Gulf economies (notably the UAE, Qatar, and Saudi Arabia), reflecting deep-rooted differences in institutional flexibility, human-capital mobility, and gender inclusion. Statistical correlation analysis highlights the emergence of a “talent ecosystem” dimension, where attraction, retention, and managerial professionalism jointly explain a large share of inter-country variance in labor-market efficiency. Egypt’s low standing (GCI 2017-2018 rank: 134) juxtaposed with the 2024 Arab frontier (UAE ranked 7th globally in overall IMD competitiveness) underscores the urgency of structural reforms in labor governance, productivity-linked compensation, and workforce adaptability. The paper contributes to the growing body of demographic-econometric literature linking labor-market efficiency to sustainable competitiveness. It offers empirically grounded insights for policymakers seeking to align human-capital strategies with regional and global competitiveness trajectories, particularly within the framework of Arab development agendas and the UN Sustainable Development Goals.

Quality Control Charts for Multivariate EWMA Daubechies Discrete Wavelet Transformation Coefficients

2026-03-17T16:31:07+00:00

Multivariate exponential weighted moving averages (MEWMA) are used to control several qualitative properties together of production processes. This article proposes the creation of new charts to control and monitor multivariate qualitative property exponential weighted moving averages, as well as variance through wavelet analysis based on the discrete wavelet transform (Daubechies). Wavelet analysis breaks down multivariate data into approximation and detail coefficients, which are used to construct the Exponential Weighted Moving Averages for approximation coefficients (MAEWMA) chart (to control and monitor the average) and the Exponential Weighted Moving Averages for detail coefficients (MDEWMA) chart (to control and monitor the variance). The proposed charts were more efficient than the conventional chart and more sensitive to slight changes, that can occur in the production processes at several values of the tuning parameter and for different sample sizes and numbers of variables through simulation studies and real data.

Analyzing the Impact of Various Criteria on Blood Pressure Through General Linear Model by Box-Cox Transformations

2026-03-17T18:10:32+00:00

Blood pressure is a key indicator of cardiovascular health, reflecting the force exerted by blood against arterial walls. Data transformation tools are essential in statistical analysis for improving assumptions necessary for linear models, such as normality, linearity, and homoscedasticity, especially when these assumptions are violated. These techniques are especially helpful for correcting distortions in data structure and enhancing the validity of General Linear Models (GLMs). In this study, we applied the Box-Cox transformation, a method from the family of power transformations, to improve a linear regression model. Our objective was to identify the most appropriate power transformation to enhance model performance and interpretability, using statistical criteria on blood test datasets collected from Azadi Hospital in Duhok. These evaluation criteria are essential for the accurate interpretation of data, as they assess the modeBlood pressure is a key indicator of cardiovascular health, reflecting the force exerted by blood against arterial walls. Data transformation tools are essential in statistical analysis for improving assumptions necessary for linear models, such as normality, linearity, and homoscedasticity, especially when these assumptions are violated. These techniques are especially helpful for correcting distortions in data structure and enhancing the validity of General Linear Models (GLMs). In this study, we applied the Box-Cox transformation, a method from the family of power transformations, to improve a linear regression model. Our objective was to identify the most appropriate power transformation to enhance model performance and interpretability, using statistical criteria on blood test datasets collected from Azadi Hospital in Duhok. These evaluation criteria are essential for the accurate interpretation of data, as they assess the model’s quality and reliability. The criteria used included: adjusted R-squared, Mean Absolute Error (MAE), Mean Absolute Percentage Error (MAPE), F-statistic, Maximum Likelihood Estimation (MLE), Root Mean Square Error (RMSE), and the Shapiro–Wilk test. These measures also guided the selection of the most appropriate GLM. We can conclude that different values optimize different model performance aspects: balances good model fit ( F-statistic). gives the most accurate predictions (lowest error). improves likelihood and residual normality. A computational algorithm was proposed to estimate the optimal power parameter, and the results of the criteria were discussed and compared. Based on the adjusted R-squared criterion, an optimal λ value was identified, indicating a strong model fit.l’s quality and reliability. The criteria used included: adjusted R-squared, Mean Absolute Error (MAE), Mean Absolute Percentage Error (MAPE), F-statistic, Maximum Likelihood Estimation (MLE), Root Mean Square Error (RMSE), and the Shapiro–Wilk test. These measures also guided the selection of the most appropriate GLM. We can conclude that different values optimize different model performance aspects: balances good model fit ( F-statistic). gives the most accurate predictions (lowest error). improves likelihood and residual normality. A computational algorithm was proposed to estimate the optimal power parameter, and the results of the criteria were discussed and compared. Based on the adjusted R-squared criterion, an optimal λ value was identified, indicating a strong model fit.

Generalized Multivariate Mixture Ratio Estimators for the Population Mean in Multi-Phase Sampling using Multi-Auxiliary Characteristics

2026-06-01T00:47:33+00:00

The integration of auxiliary information into survey sampling has continued to attract significant attention in improving the efficiency of estimators. Traditional ratio and regression estimators, though effective, often show limitations when confronted with multiple auxiliary variables and complex sampling designs. This study proposed a class of Generalized Multivariate Mixture Ratio Estimators for estimating the population mean in multi-phase sampling design with multi-auxiliary characteristics. The proposed estimators extend beyond conventional single-variable approaches by combining information from several auxiliary variables and auxiliary attributes. The theoretical properties of the estimator are derived, including the Mean Square Error expressions. Theoretical comparative analysis confirmed that the proposed estimators achieved notable gains in efficiency relative to the reviewed estimators. Simulation studies further confirmed the efficiency of the proposed estimators across varying sample sizes (asymptotically), correlation structures, and distributional conditions. Overall, the generalized multivariate mixture ratio estimators confirmed to be more efficient in population mean estimation under multi-phase sampling design.

A Comparative Study of Bayesian-Optimized Machine Learning Models for Differentiated Thyroid Cancer Recurrence Prediction

2026-03-17T19:03:58+00:00

The most prevalent type of thyroid malignancy is the differentiated thyroid cancer (DTC), and predicting its recurrence remains a clinical challenge. This study addresses the growing need for reliable predictive tools by applying machine learning techniques enhanced with Bayesian optimization. Early detection of the risk of recurrence can significantly improve health maintenance and outcomes. The study develops a comparative framework using four supervised classifiers Logistic Regression, (XGBoost) extreme gradient boost, CatBoost, and (LightGBM) light gradient boosting machine on a clinical dataset related to differentiated thyroid cancer patients. Each Model is trained and evaluated both before and after hyperparameter tuning via Bayesian optimization. Model performance is assessed using accuracy, recall, precision, and the area under the receiver operating characteristic (ROC) curve (AUC). The optimized (XGBoost) model achieved the top performance, along, recall, precision, and accuracy of 0.97, 0.99, 0.9870 respectively and an AUC of 0.9737, clearly outperforming its default counterpart. In contrast, CatBoost shows a slight performance drop after optimization, while Logistic Regression and LightGBM exhibit no significant changes. The results demonstrate that Bayesian optimization can substantially enhance model performance depending on the algorithm. This study highlights the effectiveness of optimization techniques in boosting the predictive power of machine learning models in the medical field, particularly in recurrence prediction for differentiated thyroid cancer.

A Survey on Emotion Recognition with Human Centred Software Engineering

2026-03-17T19:14:36+00:00

The survey explores the connection between emotion recognition technology and Human-Centered Software Engineering (HCSE) and, in particular, how the two aspects can be integrated to enhance user experiences, especially in customer services. The big picture is to research how emotional intelligence can be built into software system, and how this can lead to more empathic response with human beings. The work is grounded on the critical review of the current developments in the field of emotion identification, particularly on how deep learning models, namely Convolutional Neural Networks (CNNs), Recurrent Neural Networks (RNNs), and Generative Adversarial Networks are being utilized. such models are evaluated in terms of their role in real-time detection of emotional states and ratio of enhanced classification. Also, the research takes into account the multimodal recognition approaches, which involve facial expression, voice, and physiological signals to perform a thorough analysis of emotions. The survey also covers such significant implementation issues as technical constraints, ethical concerns, and algorithmic biases that exist. The results of this effort have broad implications in many fields, such as customer support, healthcare, education, and entertainment, and provide clues on how to create emotionally intelligent systems that can make the human-computer interactions much more personalized and productive.

Survival Analysis, Mortality Rate and Generalized Linear Models (GLM) with Failures (Application Study

2026-03-17T19:19:25+00:00

The number of failures of the first type is counted for each of a series of n time intervals, whereas the number of failures of the second type is only determined for the entire period. This theoretical framework is offered for the analysis of survival data when two forms of failure occur. First-type failure rates are associated with experimental and explanatory factors, while second-type failure rates are regarded as nuisance characteristics. A Latin square experiment is used to describe two models that are based on a precise and an approximate method. As long as there is a fraction of experimental units until the experiment's conclusion, the approximate model outperforms the exact model. The study presents a theoretical framework for analyzing survival data in which two types of failures occur. The first type's failures are tallied for each of n time intervals, whereas the second type's failures are only counted for the entire period. First-type failure rates are associated with experimental and explanatory factors, while second-type failure rates are regarded as nuisance characteristics.

A New Unbiased Estimator of MLR Model Coefficients Based on AC

2026-03-17T19:32:49+00:00

One of the well-known statistical methods in predictive analysis is the use of the multiple linear regression (MLR) model. In many studies, various estimators have been proposed to estimate the coefficients of this type of regression model, of which the ordinary least squares (OLS) estimator is one of the most famous and, at the same time, is one of the most accurate. This paper introduces a new estimator of MLR model coefficients based on autocovariance (AC). It is shown that although the AC-based estimator proposed in this paper may not be intuitively appealing, it is an unbiased estimator of the model coefficients. It is also shown that if the vector of independent variables satisfies certain regularity conditions, under the weak condition that the error terms follow an autoregressive moving average (ARMA) model, this estimator has the same asymptotic probability distribution as the LS estimator and converges probabilistically to the model coefficients. Finally, a simulation study confirms that the mentioned properties of the new AC-based estimator hold true even in small samples.

Adaptive Robust EWMA Control Chart for Monitoring Mean Shifts under Contaminated and Heavy-Tailed Processes

2026-03-28T16:38:21+00:00

Exponentially Weighted Moving Average (EWMA) control charts are widely employed in statistical process control to efficiently detect small and moderate shifts in the process mean. But classical EWMA models are heavily subject to normality assumptions and are very sensitive to outliers and even mild contamination in the in-control process that may lead to serious false alarm inflation and unpredictable run-length behavior. In this paper, we propose an adaptive robust EWMA control chart that can maintain stable false alarm performance under contaminated and heavy-tailed distributions. The proposed scheme uses the raw observations of EWMA recursion as a bounded robust score function with an online robust scale estimator to adjust control limits. This data-driven design ensures an approximate consistency of the in-control average run length at various levels of contamination and tail behavior. Simulations have demonstrated that the proposed chart exceeds classical and non-adaptive robust EWMA charts in terms of stability of false alarms and detection efficiency under heavy-tailed and contaminated baselines. The results indicate that the proposed adaptive robust EWMA chart is a robust and practical monitoring tool for industrial processes today with non-ideal data conditions.

Modeling Income Distributions Using Gamma and Lognormal Models: A Case Study from the University of Zakho

2026-04-03T09:07:51+00:00

Income distribution modeling plays a critical role in understanding wage structures and inequality within public institutions, particularly in developing economies such as the Kurdistan Region of Iraq. This study examines the suitability of the Gamma and Lognormal distributions for modeling the income structure of university affiliates at the University of Zakho. A systematic sample of 250 employees was selected from the university’s total workforce. Two estimation techniques Maximum Likelihood Estimation and the Method of Moments were employed to estimate the parameters of both distributions. For the Gamma model, three numerical algorithms (Newton–Raphson, Broyden, and Bisection) were applied to solve the nonlinear likelihood equations, producing stable and consistent parameter estimates. Model adequacy was evaluated using multiple goodness-of-fit tests, including Chi-square, Kolmogorov–Smirnov, Anderson–Darling, and Cramér–von Mises statistics. The results revealed that both distributions adequately describe the total salary data, with MLE providing the best overall fit (e.g., p > 0.10 in K–S and A–D tests) and lower variability compared with MEM. Conversely, both models performed poorly for basic salarie, suggesting the need for more flexible or mixed-distribution approaches in such structured data. Overall, the findings confirm the empirical relevance of distributional choice and estimation strategy in modeling income data. The study provides quantitative evidence supporting the use of MLE-based Gamma and Lognormal models as effective tools for assessing income dispersion and inequality in higher-education institutions.

Swarm Intelligence Algorithms Inspired by Nature: A Review

2026-03-17T19:26:14+00:00

This work presents a comprehensive overview of nature-inspired optimization algorithms their components, classifications, and applications in various domains. The paper focuses on the importance of basic technique in swarm optimization algorithms such as: Ant Colony Optimization (ACO), Particle Swarm Optimization (PSO), Whale Optimization Algorithm (WOA), Artificial Bee Colony Algorithm) ِABC), Salp Swarm Algorithm (SSA), Firefly Algorithm (FA), Grasshopper Optimization Algorithm (GOA), and other Swarm Intelligence Algorithms in solving complex problems. This study shows the classification of living organisms in order to classify and explore algorithms inspired by nature, and presents related works from (2020) to (2025), related to algorithms inspired by nature.