MUMBAI, India, March 13 -- Intellectual Property India has published a patent application (202641024404 A) filed by Dharun Chandru, Coimbatore, Tamil Nadu, on March 1, for 'a hybrid cnn-lstm architecture with adaptive sampling for temporal fraud pattern recognition in upi data.'
Inventor(s) include Jayakumari D S; Hemamalini S; Archana M; and Gayathiri R.
The application for the patent was published on March 13, under issue no. 11/2026.
According to the abstract released by the Intellectual Property India: "The exponential growth of Unified Payments Interface (UPI) transactions has led to increased vulnerability to sophisticated fraud schemes, making timely and accurate detection critical. This paper proposes a hybrid CNN-LSTM architecture with adaptive sampling to address temporal fraud pattern recognition in UPI data. The Convolutional Neural Network (CNN) component extracts high-level spatial features from transactional attributes, enabling identification of subtle irregularities in transaction patterns. The Long Short-Term Memory (LSTM) network captures sequential and temporal dependencies, allowing the model to detect evolving fraud behaviours over time. To mitigate the impact of severe class imbalance, an adaptive sampling strategy is introduced, which dynamically prioritizes rare and informative fraudulent transactions during training, improving model robustness and generalization. Extensive experiments on real-world UPI datasets demonstrate that the proposed approach significantly outperforms traditional machine learning algorithms and standalone deep learning models in terms of accuracy, precision, recall, and F1-score. The framework is scalable, capable of handling large transaction volumes, and suitable for real-time deployment in financial systems. This study highlights the effectiveness of integrating spatial feature extraction, temporal modelling, and adaptive sampling for intelligent and proactive fraud detection in modern digital payment platforms."
Disclaimer: Curated by HT Syndication.
