Predicting Cyber Identity Theft: A Cutting-Edge Approach using Cascaded ResNet-LSTM Deep Learning Models

Cyber identity theft has become a growing concern in the digital age, as personal and sensitive data becomes increasingly vulnerable to cyberattacks. Detecting and preventing identity theft is a critical challenge for organizations and individuals alike, requiring advanced machine learning and deep learning techniques to stay ahead of malicious attacks. In this paper, we propose a novel hybrid model, Cascaded ResNet-LSTM, for the prediction of cyber identity theft. The model integrates Residual Networks (ResNet) for spatial feature extraction and Long Short-Term Memory (LSTM) for temporal sequence learning. The combination of these two architectures allows the model to efficiently detect complex identity theft patterns by capturing both spatial and temporal dependencies in the data. The dataset utilized for this study, sourced from Kaggle, consists of 25 varied metrics and 40,000 records. Comprehensive preprocessing steps, including missing value handling, outlier detection, and dimensionality reduction using an AutoEncoder, were applied. The proposed model achieves a significantly higher accuracy of 92.65%, compared to existing models such as XGBoost (88.10%), Random Forest (86.35%), and Support Vector Machine (85.20%). Additional performance metrics, including precision (93.45%), recall (91.30%), and F1-score (92.35%), further emphasize the effectiveness of the proposed model in identifying cyber identity theft with high reliability. The hybrid architecture of ResNet and LSTM enables the model to outperform traditional approaches by reducing false positives and enhancing detection accuracy, making it a robust solution for cybersecurity applications.