Methodology of weighted fusion of heterogeneous data with an exponential obsolescence factor in financial transactions

This article develops a methodological model for fusing heterogeneous financial data in transaction risk assessment when input streams age at different speeds. The relevance of the topic follows from the spread of digital lending, instant payments, and graph-based fraud analytics, in which bureau records, behavioral traces, network signals, and contextual data enter a single decision circuit but retain unequal evidentiary value over time. The study aims to formalize a weighted fusion procedure with an exponential obsolescence factor and to define its place within auditable financial decision logic. The materials cover ten recent scholarly works on alternative data, credit scoring, fraud detection, graph learning, and multimodal fusion. Comparative analysis, conceptual synthesis, typologization, and analytical generalization were applied. The analytical section yields a source-sensitive weighting model, a recency-adjusted fusion equation, and an implementation sequence for real-time scoring and explanation. The approach applies to digital lending, payment monitoring, and anti-fraud screening.