International Journal of Data Analysis Techniques and Strategies (19 papers in press)

Regular Issues

• Enhanced Sales Forecasting Through Auto Regression and Cycle-GAN Models  
  by Arif Hossen, Md Refat Hossain, Mithun Kumar PK. 
  Abstract: Precise sales forecasting is essential for businesses to manage inventory, and allocate resources. However, traditional methods often struggle to capture sales data's complex patterns, seasonality, and dynamic nature. The problem lies in the limitations of existing forecasting techniques, which fail to model the convoluted relationships and dependencies within time series data. To address this challenge, we propose a novel method that combines the strength of autoregression(AR) and Cycle-GAN(Generative Adversarial Network) models. By applying the strengths of autoregression for capturing linear-temporal dependencies and utilising Cycle-GAN's capability to learn non-linear mappings between different-domains. Experimental results on real-world sales datasets demonstrate the excellent performance of our approach, outperforming cutting- edge forecasting methods in terms of accuracy, adaptability, and generalisation. The proposed AR-CycleGAN model delivers superior results and surpasses all other cutting-edge models with an accuracy of 98.96%, a precision of 98.16%, a recall of 98.97%, and an F1-score of 98.56%.
  Keywords: Auto Regression (AR); GAN; Cycle-GAN; Sales Forecasting; Time Series data; Machine Learning (ML); Deep Learning (DL); Business Analytics.
  DOI: 10.1504/IJDATS.2026.10073097
   
  
• Depression Detection : Analysing Social and Private Contexts for Detection with Deep Learning  
  by Gaurav Kumar Gupta, Dilip Kumar Sharma 
  Abstract: The potential social networks offer information, such as emotions, psychological behaviours, and opinions, enabling the psychological analysis to assess the mental state for depression detection. However, recognising the depression state from the linguistic content in the social network becomes insufficient. Even though social networks provide multifarious data for analysing the mindset, depression sufferers are reluctant to express their feelings publicly on social media. Thus, investigating the private context of an individual becomes crucial for accurate decision-making. Hence, considering the social and private context offers the most prominent solution to depression detection. This work proposes the social and private context-based depression (SPriD) detection model using deep learning. Moreover, the proposed approach integrates the depression tendency from social and private contexts to distinguish the depressive and non-depressive individuals. Thus, the results of SpriD show the superiority of the proposed depression detection approach.
  Keywords: Depression Detection; Social Context; Private Context; NRC lexicon; Word-level Weighted Vectorization; Multi-Task Semi-Supervised Learning; Weighted attention; and Hybrid Deep Learning.
  DOI: 10.1504/IJDATS.2026.10073294
   
  
• Machine Learning Algorithms to Predict Groundwater Productivity in Iraq  
  by Qahtan Yas, Younis K.Hamead 
  Abstract: Groundwater (GW) is a vital water source in most countries, but the decline in surface and groundwater supplies is a significant challenge. This paper presents a promising solution by proposing a new approach to predict future groundwater levels using machine learning algorithms. The dataset of groundwater productivity for the period (20172020) was adopted in Iraq. The methodology was implemented to compare and evaluate the performance of four different machine learning models: the Elman Neural Network, Cascade Neural Network, Layer Recurrent Neural Network, and Nonlinear Autoregressive with Outlier for predicting groundwater productivity in Iraq. Various metrics were adopted to evaluate the proposed ML algorithms. The results showed that the NARX algorithm performed the best in predicting groundwater productivity at the value 4804.68, outperforming other models. This research has the potential to significantly impact future water resource management, offering hope for a more sustainable future.
  Keywords: Groundwater Productivity; ENN algorithm; LRNN algorithm; CNN algorithm; NARX algorithm; Machine Learning algorithms; Sustainability.
  DOI: 10.1504/IJDATS.2026.10074703
   
  
• Drawing the Profile of the Parent's Intention to Control their Child's Internet Use with the Naive Bayes Method  
  by Esin Avci 
  Abstract: The rapid growth of internet access among children brings both opportunities for learning and significant exposure to online risks. This study applies the Naive Bayes classification algorithm to identify factors influencing parents decisions to install special software designed to block access to harmful websites. Data were collected via surveys from parents, teachers, and students in 26 schools in Giresun, Turkey, encompassing demographic, occupational, and household characteristics. Results indicate that only 15.5% of parents use such software, with adoption more prevalent among parents aged under 35, those with two children, and those employed in the state sector or not currently working. The Naive Bayes model achieved an accuracy rate of 89% (CI: 8195%), demonstrating strong predictive capability for software installation behaviour. These findings underscore the importance of targeted awareness and educational programs for parents, particularly in promoting digital safety measures for children.
  Keywords: Internet security; children; software; machine learning; Naive Bayes classifier.
  DOI: 10.1504/IJDATS.2026.10074711
   
  
• Identification of the Importance Level of Characteristic Variables for Food-Insecure Households Using Random Forest  
  by Muhammad Subianto, Riska Adelia, Nany Salwa, Evi Ramadhani, Bagus Sartono 
  Abstract: The rise of big data presents both challenges and opportunities for machine learning applications. Random Forest, an ensemble method using Decision Trees and bagging, offers strong predictive performance but often lacks interpretability. This study aims to develop an optimal classification model and identify key factors associated with food-insecure households. Data were sourced from the 2022 National Socioeconomic Survey (SUSENAS) in Aceh Province. The best model used a 55:45 data split with optimised hyperparameters: 88 estimators, 47 max features, 37 max depth, min samples split of 5, min samples leaf of 1, and the entropy criterion. Model performance reached 65.88% accuracy, 70.50% precision, 65.88% recall, and 67.83% F1-score. SHAP was applied to interpret model outputs, revealing the five most influential variables: residence floor area, adequacy of sanitation facilities, education of household head, land asset ownership, and internet access.
  Keywords: Classification; Random Forest; SHAP; Food Insecurity; Aceh.
  DOI: 10.1504/IJDATS.2026.10074894
   
  
• Making Data Visualisation more Efficient and Effective  
  by Rania Mkhinini Gahar, Olfa Arfaoui, Minyar Sassi Hidri 
  Abstract: Data science is an interdisciplinary study field that uses systems, processes, algorithms, and other frameworks tomake use of massive data amounts. Thus, data scientists integrate a variety of abilities, such as IT, statistics, and business knowledge to evaluate data gathered from clients or other sources utilising sensors, smartphones, web surfing patterns, etc. However, it is their still little-known competence that makes their profile so appealing to recruiters. This is why these uncommon profiles are in high demand. This paper is a return on a definitely attractive and constantly evolving data scientist profession in terms of dashboards highlighting its salary. Defining dashboard performance indicators for data scientists and designing a management and reporting tool which is, to be served for data enthusiasts to rush into the Data Science world and this by using cutting-edge technologies are the main contributions of the proposed work.
  Keywords: Data Science; DataViz; Business Intelligence; Power BI; Dashboard; Big Data.
  DOI: 10.1504/IJDATS.2026.10075121
   
  
• A Novel Algorithm for Tree-Based Sequential Pattern Mining using HU-Chain structure  
  by Ritika ., Sunil Kumar Gupta 
  Abstract: One of the challenges faced by researchers in the domain of mining high-utility sequential patterns is maintaining the downward closure feature. Addressing this issue, researchers have proposed various properties, representations, and pruning strategies to be utilized in high-utility mining. This paper proposes a novel compact header utility chain data structure for representing the utility information providing quick access to utility values while traversing the lexicographic quantitative sequence tree. The candidates are effectively reduced using a combination of pruning strategies. The sequence identifier information is stored only once in the header node during the construction of utility chain thereby saving time. Experimental data shows that the proposed approach performs better than the current approaches in terms of mining speed as well as the number of candidates created.
  Keywords: Header utility chain; pruning strategies; tree; high utility mining.
  DOI: 10.1504/IJDATS.2026.10075753
   
  
• Predicting Academic Performance in Computational Sciences: Utilising Naive Bayes and SVM Models with Student Interest and Course Data  
  by Ikpotokin Osayomore, Lekia Nkpordee, Yusuf Abass Aleshinloye, Osasumwen Usen 
  Abstract: This study aimed to predict academic performance in computational science courses using machine learning models, comparing the effectiveness of Naive Bayes and Support Vector Machine (SVM). A dataset of student exam scores and interest levels in four courses was analysed. Results showed that Naive Bayes consistently outperformed SVM in predicting student performance. Forecasts using Naive Bayes indicated a positive academic trend, particularly in Data Analysis and Simulation & Modelling, with expected improvements of 7-10% in future sessions. The study highlights Naive Bayes as a reliable tool for early identification of at-risk students, recommending its use in academic monitoring systems to enhance personalised learning interventions. Future work could explore hybrid models to further improve prediction accuracy.
  Keywords: Machine Learning; Educational Analytics; Performance Forecasting; Student Engagement; Predictive Modelling.
  DOI: 10.1504/IJDATS.2026.10076100
   
  
• Empirical Study of WeakGamma and WeakVQRS Fuzzy Measures for Fuzzy Rough Feature Selection  
  by Andreja Naumoski 
  Abstract: This paper explores the influence of two fuzzy measures on the feature selection process using fuzzy rough set theory (FRST) combined with advanced search strategies, including Ant Colony Optimization (ACO). As a part of the model performance evaluation process, we evaluate and compare full-feature models and fuzzy rough feature-selected models across sixteen diverse datasets using variety of classification algorithms. The FuzzyRoughSubsetEval evaluator is utilized with both WeakGamma and WeakVQRS fuzzy measures to assess the relevance of feature subsets. Model performance is assessed by the area under the receiver operating characteristic curve (AUC-ROC), and statistical significance is determined through a two-stage nonparametric testing procedure. The results show that fuzzy rough feature selection, particularly when leveraging ACO and different fuzzy measures, can produce models with competitive generalization compared to full-feature models. The choice of fuzzy measure is shown to be important for optimizing both accuracy and robustness.
  Keywords: Fuzzy Rough Set Theory; Feature Selection; Ant Colony Optimization; Fuzzy Measures; AUC-ROC.
  DOI: 10.1504/IJDATS.2027.10076239
   
  
• Data Quality Failures and Symbolic Collapse in Business AI Systems: a Qualitative Modelling Approach  
  by Ruben Xing, David Eisenberg 
  Abstract: As artificial intelligence (AI) systems become central to enterprise decision-making, their most disruptive failures often arise from data quality breakdowns and governance gaps rather than model design flaws. This study examines how inconsistencies, semantic drift, and fragmented escalation pathways propagate in real-world deployments, drawing on 32 field interviews across five industries. Through qualitative synthesis, we identify a recurring failure cascade spanning three layers: input contamination, reasoning misalignment, and delayed remediation. We contribute two governanceoriented frameworks that integrate diagnosis and prevention. First, a failurecascade model traces disruption patterns and symbolic thresholds at which reasoning collapses, including spurious correlation dominance, critical signal-to-noise boundaries, and drift recovery windows. Second, an AI readiness matrix maps organisational roles and escalation responsibilities to enable timely intervention. By combining qualitative cases with symbolic scaffolds, the study advances an evidence-based framework for resilient and auditable AI, emphasising proactive quality management over post-hoc model tuning.
  Keywords: Artificial intelligence; data quality; failure cascade; organizational risk; qualitative analysis; AI governance; machine learning breakdowns.
  DOI: 10.1504/IJDATS.2026.10076249
   
  
• Financial Market Prediction Using Ensemble Deep Learning Meta-Algorithms  
  by Nesa Sadeghi, Jaber Fooladi, Nasser Ghaem Doust 
  Abstract: In volatile financial markets, predictive systems must balance return and risk effectively. This study introduces an Autoencoder-based Ensemble Learning framework that extracts latent market features and reduces dimensionality, addressing complexity and noise in financial time series. By distributing learning across multiple base predictors with randomized weighting, the ensemble adaptively balances variance, bias, and overfitting, achieving robust generalisation. Evaluated on the S&P 500 index, the model outperforms traditional statistical and machine learning approaches in both absolute and risk-adjusted returns. Its adaptive variance management and portfolio diversification mechanisms enable stable, consistent performance, even under turbulent market conditions. For asset managers, hedge funds, and risk management teams, the framework provides actionable insights for portfolio allocation, volatility control, and proactive risk mitigation. This study demonstrates that combining unsupervised deep learning with ensemble strategies not only enhances predictive accuracy but also transforms complex financial data into practical, decision-driven business value.
  Keywords: Ensemble Learning; Deep Learning; Autoencoder; Bagging; Sharpe Ratio.
  DOI: 10.1504/IJDATS.2026.10076343
   
  
• The Evolution of Technology Innovations and Implications for Sport Management  
  by Minh-Hieu Le, Phung Phi Tran 
  Abstract: This manuscript presents a comprehensive bibliometric analysis of technological innovations in the sports industry from 2014 to mid-2024. Utilizing bibliographic coupling and keyword co-occurrence methods, complemented by network and overlay visualizations, the study examines author productivity, publication trends, and the evolution of key research themes. The analysis reveals a clear temporal shift in scholarly focus. In particular, the period from 2020 onward is marked by significant advancements in performance tracking technologies and a notable emergence of research in cutting-edge areas such as artificial intelligence and blockchain. The keyword co-occurrence analysis identifies seven major thematic clusters, encompassing advanced computing technologies, user engagement, health and injury management, and performance analytics. Network and overlay visualizations further underscore the dynamic progression of research interests, highlighting the increasing prominence of AI and blockchain-driven innovations in the contemporary sports technology landscape.
  Keywords: Technological Innovations; Sports Industry; Bibliometric Analysis; Artificial Intelligence; Performance Tracking Technologies.
  DOI: 10.1504/IJDATS.2027.10077566
   
  
• Behavioral Entropy in Daily Life: An Integrated Machine Learning Strategy for Time Allocation Analysis  
  by Jack Lesser 
  Abstract: Entropy, originally from thermodynamics and formalised in information theory by Shannon (1948), measures uncertainty or diversity in a system. Applied to daily time allocation, entropy captures how individuals distribute minutes across activities. High entropy indicates time spread across many activities; low entropy indicates concentration into few. This study examines predictors of individual differences in behavioural entropy using data from the American Time Use Survey (ATUS) (N = 18,260). An integrated machine learning pipeline combining LASSO regularisation, Random Forest validation, and weighted regression identifies which activities and characteristics are associated with higher or lower entropy. Findings reveal that household management and childcare responsibilities drive high entropy, while extended sleep and television viewing are associated with lower entropy. Consumer activities (shopping) contribute minimally. The pipeline demonstrates a replicable strategy for analysing high-dimensional time-use data when theory provides limited guidance on variable selection.
  Keywords: behavioral entropy; machine learning; LASSO; Random Forest; time allocation; daily activities.
  DOI: 10.1504/IJDATS.2027.10077639
   
  
• A Bibliometric Systematic Literature Review on Machine Learning Methods for Predicting Customer Churn  
  by Irfan Saleem, Farzad Fallah, Syed Muhammad Ali Shahbaz Habib, Abdul Rauf 
  Abstract: Customer churn, the termination of a customer's relationship with a business, is a significant challenge in the business domain, as the cost of acquiring customers through marketing practices is much higher than retaining existing customers. Advancements in machine learning (ML) have enabled new methods for predicting customer churn risk using large datasets. This paper reviews ML methods for churn prediction, including supervised, unsupervised, and other advanced approaches, and explores their strengths and weaknesses, as well as their practical applications, through a bibliometric systematic literature review. Key methods discussed include logistic regression, decision trees, support vector machines, clustering techniques, neural networks, and feature engineering and selection. The study also evaluates these methods using metrics such as accuracy, precision, recall, and F1-score to measure their effectiveness in churn prediction. Additionally, this research applies three ML methods. The study also recommends future research for scholars and provides implications.
  Keywords: Machine Learning; Customer Churn; Logistic regression; Decision tree; K-nearest neighbors; Support Vector Machine; Bibliometric Systematic Literature Review.
  DOI: 10.1504/IJDATS.2027.10077898
   
  
• Segment-Specific Consumer Risk in Online Reviews A Dual Sentiment and Fuzzy Logic Approach  
  by Alp Par 
  Abstract: Understanding consumer perceived risk in health related product categories requires approaches beyond aggregate satisfaction metrics. This study examines segment specific risk perceptions by integrating sentiment based text analysis with a fuzzy Failure Mode and Effects Analysis framework. Using a large scale dataset of consumer skincare reviews, risk dimensions related to dryness, irritation, breakout, and scent are identified. Severity, occurrence, and detectability are operationalized through rating patterns and linguistic signals to compute fuzzy Risk Priority Numbers at the segment level. The results reveal substantial heterogeneity across skin types, indicating that adverse outcomes are not uniformly experienced or perceived. Certain risk dimensions emerge as salient for specific segments while remaining secondary for others. These findings highlight the limitations of population level risk indicators and demonstrate the value of segment sensitive risk assessment for health marketing decision making.
  Keywords: Consumer risk; Online reviews; Dual sentiment; Fuzzy logic; Risk prioritization; Segment heterogeneity; Skincare analytics.
  DOI: 10.1504/IJDATS.2027.10078998
   
  
• From Data to Prada: Automated Generation of Editorial Trend Narratives from Social Media Using Large Language Models  
  by Sarah Bouraga 
  Abstract: Social media platforms have become valuable sources for understanding consumer preferences and identifying emerging trends. However, converting raw social media data into actionable insights requires significant manual effort from domain experts. Recent advances in Large Language Models (LLMs) offer potential for automating aspects of this process, particularly the generation of human-readable trend narratives. This paper presents a practical system that integrates traditional Natural Language Processing techniques with LLMs to automatically extract and articulate trends from Instagram influencer content. We collected 8,758 Instagram captions from 20 top fashion influencers covering September 2022 to August 2023. Our system applies text preprocessing, TF-IDF vectorization, embedding-based clustering, and topic modeling to identify thematic groups, then leverages ChatGPT and Gemini through zero-shot prompting to generate editorial-style trend descriptions.
  Keywords: Social Media Analytics; Data Analysis Techniques; Text Mining; Fashion Industry Applications;Trend Detection; Large Language Models.
  DOI: 10.1504/IJDATS.2027.10079170
   
  
• Alternative Data in Credit Scoring: Quantifying the Information Loss Using Kullback-Leibler Divergence  
  by Edwin Baidoo, Ramachandran Natarajan 
  Abstract: Alternative datasets are increasingly used in finance due to their convenience and availability. In applications such as lending, alternative data promotes financial inclusion. The consequences of such use are not always clear. Therefore, exploring the effects of alternative data offers rich research opportunities. In this paper, we use Kullback-Leibler Divergence to quantify information loss when alternative datasets are used to make lending decisions. We formulate hypotheses and test the statistical significance of the loss. Additionally, we address the economic and operational significance of the loss and its impact on lending decisions. The paper suggests that alternative data should be evaluated for information loss, and financial inclusion considerations should be balanced against the loss of information quality.
  Keywords: Alternative Data; Kullback-Leibler Divergence; Credit Scoring and Lending; Financial Inclusion.
  DOI: 10.1504/IJDATS.2027.10079243
   
  
• Directional Decision Tree with Absolute Unique Median Filter for Effective Impulse Noise Removal  
  by Rasyida M.D. Saad, Ahmad Kadri Junoh, Wan Zuki Azman Wan 
  Abstract: Impulse noise, especially salt and pepper noise, significantly degrades image quality and affects the accuracy of image-based analysis. This study proposes a new approach, Directional Decision Tree with Absolute Unique Median Filter (DDTAUMF), for denoising grayscale images. The approach involves two main stages: database creation and image restoration. Singular Value Decomposition (SVD) is used to detect and localize noise, which is then classified into 18 directional classes based on pixel coordinates. In the restoration phase, a decision tree trained across 9 directions guides the identification of noisy pixels, which are then restored the new pixels using an Absolute Unique Median Filter. Experimental results on nine standard grayscale images with noise densities ranging from 10% to 90% show that DDTAUMF outperforms existing methods in terms of Peak Signal-to-Noise Ratio (PSNR) and Structural Similarity Index Measure (SSIM), particularly under high noise levels, while preserving critical image details.
  Keywords: Image denoising; Impulse noise; Singular value decomposition; Decision tree; Median filter.
  DOI: 10.1504/IJDATS.2027.10079767
   
  
• Perception Analysis about Drug Tweets: A Performance Analysis of Different Approaches  
  by Radhakrishna Jana, Dharmpal Singh, Saikat Maity, Ankush Mallick, Soumita Seth, Somenath Chakraborty, Saurav Mallik 
  Abstract: On social networking sites like Facebook and Twitter, a very large amount of data of information have been found that contains users or customers feelings and emotions in the form of reactions and comments to tweets in Twitter. These data are very helpful for finding out the overall sentiment for any type of posts or tweets on Twitter and with the help of Sentiment Analysis (SA), whether it corresponds to positive sentiment or negative sentiment. In this research work, sentiment analysis is done for the tweets for any type of drugs. Here, seven different machine learning models are used for evaluating the sentiment of each tweet. The two datasets are used in this research work is taken from the UCI Machine Learning Repository. By applying these seven machine learning models, the LightGBM Classifier (LGBM Classifier) performs best with a testing accuracy of 90.78%.
  Keywords: Sentiment Analysis; LightGBM; XGBoost; CatBoost; Gradient Boosting Classifier; Random Forest; K-Nearest Neighbors; Logistic Regression; Drug Reviews; Drug Tweets; Twitter.
  DOI: 10.1504/IJDATS.2027.10079768