Terahertz PCF sensor for explosive detection: A machine learning approach to nitroglycerine and royal demolition analysis

• High Sensitivity Detection: The proposed square hollow-core PCF sensor achieves exceptional relative sensitivity—up to 98.09% for Nitroglycerin and 88.25% for RDX—within the 1 to 2.8 THz range. • Low Loss and Broad Effective Area: Optimized structural parameters ensure minimal effective material loss and a large effective area, enhancing the sensor’s overall optical performance. • FEM-Based Numerical Analysis: The sensor’s performance is validated using Finite Element Method (FEM) simulations, confirming its precision in explosive detection. • Machine Learning Integration: A Random Forest Regressor is employed to predict sensor behavior, achieving a perfect R² score and enabling accurate forecasting of parameters like sensitivity and EML. • Fabrication Readiness: The sensor design is compatible with current fabrication technologies, making it practical for real-world deployment in security and safety applications. This article presents a square hollow core Photonic Crystal Fiber (PCF) sensor developed for high relative sensitivity detection of explosives (Nitroglycerin and Royal Demolition Explosive (RDX)) in the terahertz region (1 THz to 2.8 THz). The numerical sensing capabilities are assessed utilizing the finite element technique(FEM). We have attained enhanced relative sensitivity with negligible loss for detecting Nitroglycerine and RDX through the optimization of structural factors. The maximum relative sensitivity achieved is 98.09 % for Nitroglycerine and 88.25 % for RDX at 2 THz. Additionally, we have achieved little effective material loss (EML) and an extensive effective area. The proposed sensor design is compatible with current fabrication technologies, ensuring practical feasibility. Furthermore, the prediction was conducted with the Random Forest Regressor. We have attained optimal accuracy of prediction with a unity R 2 score, and this model may be utilized for predicting much behaviour, including relative sensitivity and EML for frequency.