Modelling An Iot-Enabled Shapley Value Pattern Analysis and Transfer Learning Model for Women Security Enhancement
TL;DR
The suggested IoT-driven women security framework, combining a DNN with transfer learning and SHAP-based explainability, achieved an overall accuracy of 96.00% on the target-domain dataset, and the transfer-learning strategy reduced training time by approximately 38% and memory consumption by 80% compared with training from scratch.
Abstract
IoT enhances governance through analytics of data and continuous surveillance in women's security, and AI simplifies processes and boosts production. However, because to issues including domain differences, complexity of computation, biased sample selection, negative transmission, and a lack of data compared with inscription, AI technologies need a significant amount of security information and training time. Additionally, it is difficult for human operators to comprehend and trust AI decisions due to the unreliable nature of IoT over AI patterns. To overcome these obstacles, this investigation suggests an IoT-based security administration platform for women security that makes use of cutting-edge technology to improve operational effectiveness as well as security. The recommended approach uses robust Deep Learning patterns to ensure transparency and reliability in personal security administration created by learning approaches for analyzing the patterns and enhanced with XAI. The creation of an extensive IoT-based personal security framework, a thorough case investigation to maximize DL pattern effectiveness employing DNN with shapley values for analyzing the dataset patterns, and the primary contributions of the current study are the creation of a laboratory for thorough validation. The suggested IoT-driven women security framework, combining a DNN with transfer learning and SHAP-based explainability, achieved an overall accuracy of 96.00% on the target-domain dataset. Based directly on the confusion matrix, the safe class obtained 98.94% precision, 93.00% recall, and a 95.88% F1-score, whereas the unsafe class obtained 93.40% precision, 99.00% recall, and a 96.12% F1-score. The macro-averaged precision, recall, and F1-score were 96.17%, 96.00%, and 96.00%, respectively, with an MCC of 0.9217. Furthermore, the transfer-learning strategy reduced training time by approximately 38% and memory consumption by 80% compared with training from scratch, supporting deployment on resource-constrained IoT edge devices.