Journal of Big Data Analytics and Business Intelligence

Customer-Level Differential Privacy in Federated Repeat Purchase Prediction with Unequal Retail Clients

Tue, 27 Oct 2026 00:00:00 +0000

Federated customer prediction requires a clear distinction between decentralised optimisation and a formal privacy mechanism. This study implements customer-level clipped gradient aggregation for repeat-purchase prediction using UCI Online Retail. A fixed 90-day history precedes 1 November 2011, and the outcome records purchases during the following 30 days. A deterministic customer split produces 1,457 training, 497 validation, and 507 test observations. Countries define 27 unequal simulated training clients, with the largest containing 1,316 customers. Each customer contributes one bounded feature vector and one clipped logistic gradient per round. Gaussian noise is added to the aggregate under a trusted coordinator, and a Rényi accountant reports replacement-adjacency privacy bounds for 100 full-participation rounds. Nonprivate federation achieves test ROC AUC of 0.6769 and average precision of 0.6366. At a per-model privacy bound of = 3.194 and = 0.00001, mean ROC AUC is 0.6812 and average precision is 0.6397 across five noise seeds. This small numerical increase is not evidence that privacy improves prediction. The results demonstrate a reproducible privacy-utility calculation for a limited retail cohort, without establishing distributed system scalability or protection against an untrusted aggregation server.

Multimodal Machine Learning for Crop Yield Prediction in Precision Agriculture

Akshaya Uttekar, Dattatraya Kumbhar, Ashwini Kumbhar — Thu, 17 Sep 2026 00:00:00 +0000

Increased focus on precise agricultural insurance, climate-risk management, food-security planning, and precision agriculture has led to a greater interest in accurate prediction of crop yield. The development of Earth observation, meteorological reanalysis, soil mapping, field sensing, and Machine Learning (ML) has moved yield forecasting beyond single-source statistical models to multimodal predictive models. This critical analysis covers the emerging trends in ML and Deep Learning (DL) to predict crop yields, with a particular focus on the combination of remote-sensing images, weather patterns, soil and topographical data, crop-management data, and crop-growth indicators. Random forest, gradient boosting, convolutional neural networks, recurrent networks, and hybrid process-ML are effective predictors in representative studies; however, reported performance is highly dependent on the type of crop, the scale of the spatial area, the quality of the data, representation of the features, and the approach to validation. The multimodal integration can be specifically beneficial due to the description of complementary aspects of the crop-production system by different data sources, but the mismatch in spatial resolution, uncertainty in ground truth, leakage of information, poor geographic transferability, and poor model interpretability are the critical issues. New directions are process-guided learning, explainable AI, transfer learning, foundation models, uncertainty-aware forecasting, and computationally efficient operational systems. Future developments should hence be assessed in both predictive accuracy and robustness, reproducibility, agronomic plausibility, and practical decision value.

Data Mining and Business Forecasting: Enhancing Decision-Making through Intelligent Data Analytics

Rahul Kumar, Ritesh G. Upadhyay — Thu, 17 Sep 2026 00:00:00 +0000

The sheer volume, variety, and complexity of commercial information demand improved analytical approaches that can support accurate forecasting and timely managerial planning. This work presents a hybrid data mining/business forecasting approach that is developed to extract hidden patterns from complex information sets and increase the predictive accuracy of business prediction. It integrates data pre-processing, variable selection, clustering, nonlinear learning, and business forecasting techniques to address challenges in data quality, high dimensionality, nonlinear relationships, and temporal dependencies. The system can use classification, regression, association rule mining, neural networks, ensemble learning, gradient boosting, LSTM-based forecasting, and time-series analysis techniques, with hybrid models combining different models to complement each other and make the best use of each. The proposed setup is based on data pre-processing, conceptualization, clustering and association analysis, predictor selection, model building and validation, comparison with traditional individual modelling, and the use of suitable forecasting and classification performance measures. The setup seems to assure better, more accurate, and more reliable business prediction compared to a standard, single course of action. The feature selection and hybrid modelling reduce the effects of redundant and non-representative information, reducing the model complexity. Empirical application has demonstrated the capacity of the system to improve forecasting, customer mining, risk calculations, and many other data mining-based business processes. The general system can be seen as a practical approach to the data analysis and business forecasting domain. It offers flexibility and reliability that may contribute to improving demand forecasts, reducing uncertainty and bidding risk, and increasing the operational efficiency of the institution.