Gundeti Mounika, Vimula Shwetha, Dr K Sridhar · JEE 2026 · 2026
DOI: 10.65713/jeev2i401
Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).
Organizations increasingly use data-driven systems to support operational decision-making in dynamic environments. However, many existing deep learning models focus primarily on predicting outcomes rather than recommending reliable actions. These models often provide limited information about prediction confidence, may experience performance degradation when data patterns change, and offer insufficient explanations for their decisions. Previous studies have addressed these challenges independently through improved prediction models, uncertainty estimation, calibration techniques, data drift detection, and explainable AI methods. To address these limitations, this paper introduces IntelliLayer, a five-layer deep learning architecture that combines these capabilities within a unified decision-making framework. The proposed architecture integrates static data and sequential events using Transformer encoders and a learnable gated fusion mechanism. It estimates calibrated outcome distributions through a deep ensemble and selects actions based on expected utility, while uncertain cases are referred to human reviewers. To handle changing data patterns, the framework employs adaptive windowing for drift detection and updates the model using replay techniques and elastic weight consolidation. Additionally, integrated gradients are used to explain individual predictions, and an audit record is maintained for every decision to support transparency and accountability. The paper describes the architecture of each layer, outlines the online decision-making algorithm, and presents a reference implementation stack. It also defines a reproducible evaluation framework covering predictive performance, calibration, decision-making costs, selective prediction coverage, recovery from data drift, and processing latency. Since experimental results have not yet been obtained, the proposed evaluation protocol and results template provide a structured basis for future validation of the IntelliLayer architecture.
No comments yet — start the discussion below.