Journal of Knowledge in Data Science and Information Management

Image Detection Using Convolutional Neural Networks: A Review of Transfer Learning, Ensemble Learning and Gated Expert Models for AI-Generated Image Detection

Pallavi Tukaram Chaudhari — 2026-10-30

Image detection and classification are the core problems in computer vision, as demonstrated by the emergence of Convolutional Neural Networks (CNNs), which enable learning visual representations directly from images without requiring manual engineering of descriptors. Consequently, CNNs can be leveraged to solve practical image recognition tasks, including the detection of synthetic and AI-generated content with a high degree of accuracy. This review article provides an overview of the image detection problems, focusing on the CNN-based approach. The article reviews the CNN developments and outlines the transfer learning approach that was employed to fine-tune the pre-trained ResNet-50 convolutional network for the task of image detection. Additionally, the study discusses the approaches that employ ensembling, such as averaging, voting, unified training, gated mixture-of-experts, and the application of multiple CNNs to solve the detection problem stated. Finally, the review compares the findings with other related works, noting the trends that reveal the difficulties of developing a reliable and general image detection method, particularly in terms of performance differences across different types of generators. The computational costs and limitations of the current approaches are also described alongside potential research directions, which include the development of larger and more comprehensive benchmark data sets for training and testing detection algorithms. The study concludes by stating that while CNNs are an effective tool for image detection due to their capacity to represent high-level concepts with relatively low computational cost, the generalization ability of current methods is often insufficient to address real-world problems, and thus ongoing research efforts are required.

An AI-Driven Multi-Criteria Decision Support Framework for Explainable Decision-Making under Uncertainty

Shaik Humera — 2026-10-07

Decision-making in the context of modern organizations can be regarded as a complex process due to multiple factors, data volumes, contradicting criteria, uncertainty, and dynamic conditions. Therefore, it is essential to consider advanced methods for analyzing, prioritizing, comparing, and selecting alternatives. This study aims to propose an idea for an AI-based Multi-Criteria Decision Support Framework for Intelligent Decision-Making in Dynamic Environments based on the principles of artificial intelligence, machine learning, Multi-Criteria Decision Support, fuzzy logic, data management, and visualization techniques. The suggested framework could be considered a structure supporting the process of decision-making through analyzing alternatives on multiple criteria and providing recommendations. The programming language Python could be used as an efficient environment for implementing numerical calculations, data science, machine learning, and visualization. In particular, NumPy and Pandas are efficient tools for mathematical functions, statistics, and working with data, while Scikit-learn and SciPy provide machine learning algorithms, optimization, and statistical functions. Moreover, the Python language has libraries for creating visual content for further presentation and exploration of results. The framework could implement machine learning algorithms such as Logistic Regression, Support Vector Machine, k-Nearest Neighbors, Random Forest, and XGBoost for classification, prediction, risk assessment, and pattern recognition. The Analytic Hierarchy Process, TOPSIS, VIKOR, and EDAS methods could be applied for multi-criteria analysis of alternatives, while fuzzy logic could be used for representing uncertainty in natural terms.

Evaluating Transformer and Graph Aggregation for Explainable Retail Product Growth Prediction

Shailendra Singh Tomer — 2026-10-26

Multimodal prediction should demonstrate complementary information rather than assume that a more complex architecture discovers better knowledge. This study evaluates a small Transformer and fixed graph aggregation hybrid on linked UCI Online Retail records. Historical product descriptions form the text input, six sales-history measurements form the numerical input, and shared customers define a product graph. Using 90-day histories at the September, October, and November 2011 cutoffs produces 1,983 training, 2,137 validation, and 2,273 test products. The target records whether positive units sold in the next 30 days exceed those in the preceding 30 days. Graphs, vocabulary, and inputs respect the relevant observation cutoff, while validation average precision selects among three training checkpoints. Across five seeds, the numerical model obtains test average precision of 0.6923 and ROC AUC of 0.6881. The full hybrid obtains average precision of 0.5658 and ROC AUC of 0.5331, while the numerical-plus-graph model obtains average precision of 0.6793. Modality occlusion examines fitted-model sensitivity but is not treated as causal explanation. The results do not support hybrid superiority in this limited temporal benchmark and show why a linked multimodal representation requires empirical comparison with simple historical features.