MUMBAI, India, Oct. 12 -- Intellectual Property India has published a patent application (202641118499 A) filed by Mothukuri Sai Prasad Rao; Dr. Chandra Srinivas Potluri; Dr. Sreedevi Kadiyala; Ravindra Changala; and Pillalamarri Laxman on October 04, 2026, for A Machine Learning Enabled Modified Lgbm Approach For Efficient Iot Malware Detection.
Inventors include Mothukuri Sai Prasad Rao; Dr. Chandra Srinivas Potluri; Dr. Sreedevi Kadiyala; Ravindra Changala; and Pillalamarri Laxman.
The application for the patent was published on October 09, 2026, under issue no. 41/2026.
Abstract: ABSTRACT OF THE INVENTION [067] A machine-learning-enabled system and method for efficient Internet of Things (IoT) malware detection utilizing a modified Light Gradient Boosting Machine (LightGBM) approach is disclosed. The system comprises a network capture module for intercepting IoT traffic, a feature extraction module for deriving relevant features from captured packets, and a correlation-based feature selection module for identifying discriminative features while eliminating redundant attributes, thereby reducing feature dimensionality by up to 70 percent. [068] The detection engine comprises a modified LightGBM classifier trained with an optimized sample-weighting scheme employing a Reduced Churn Operator (RCOP) that assigns weights based on historical classification performance. Samples misclassified by previous model versions receive reduced weights, while correctly classified samples receive higher weights. This methodology enables the model to maintain detection accuracy exceeding 99 percent while substantially reducing bad churn compared with conventional retraining approaches. [069] An incremental- learning module enables rapid adaptation to emerging malware threats without complete system retraining, employing Gaussian mixture models and density-based clustering to identify and incorporate new malware domains. Experimental evaluation on the IoT-23 dataset demonstrated 99.2 percent accuracy, 0.09 percent bad churn, 42 percent reduction in processing time, and inference latency of 4.3 ms per sample on low-power ARM processors. The lightweight inference engine requires only 124 KB of memory and 15.6 MB of storage, making it suitable for edge deployment. [070] The present invention addresses the limitations of existing IoT malware-detection systems by providing a computationally efficient, accurate, and adaptable solution with consistent detection performance across successive model updates. The integration of correlation-based feature selection, modified LightGBM classification with RCOP sample weighting, and incremental-learning mechanisms represents a significant advancement in machine-learning-based malware detection for resource-constrained environments, finding application in cybersecurity systems that protect IoT devices against evolving threats including ransomware, botnets, trojans, and zero-day attacks.
Disclaimer: Curated by HT Syndication.
