CINN-UTLC: A Computationally Intelligent Neural Network-Based Unsupervised Transfer Learning Algorithm for Ransomware Detection

The relentless evolution of ransomware demands detection frameworks that adapt to novel variants and minimize reliance on labelled data.Existing methods often suffer from distribution shifts, limited generalizability, and opaque decision-making.This study introduces CINN-UTLC, a computationally intelligent neural network-based unsupervised transfer learning algorithm that integrates domain adaptation, hybrid feature extraction, and explainable clustering for ransomware detection.By combining Convolutional Neural Networks (CNNs) and Long Short-Term Memory (LSTM) models, CINN-UTLC captures static features (e.g., file entropy, headers) and dynamic behaviours (e.g., API call sequences) while aligning source (benign) and target (unlabelled) domains via Geometric Alignment Clustering (GAC).The framework employs SHapley Additive exPlanations (SHAP) and Local Interpretable Model-Agnostic Explanations (LIME) to interpret feature contributions, ensuring transparency in clustering decisions.CINN-UTLC achieves a 98% detection rate, 2.5% false positive rate, and AUC of 0.94, outperforming benchmarks like UNVEIL (AUC=0.78)and deep learning methods (AUC=0.73).Clustering metrics (Silhouette Score: 0.80-0.86;Adjusted Rand Index: 0.87-0.93)confirm robust separation of ransomware families, including zero-day variants.The algorithm's unsupervised transfer learning capability enables detection of unknown ransomware through behavioural anomalies, even without labelled target data.By addressing domain shifts, reducing false positives, and offering explainable insights, CINN-UTLC sets a new standard for adaptive cybersecurity frameworks, bridging critical gaps in ransomware resilience and proactive threat mitigation.