Accurate miniaturized representation of silicon-on-insulator (SOI) MOSFETs cannot be avoided when on the path circumference designs of general circuits are required. Simulations done using physics give high fidelity, but also have a high computational cost reducing their usefulness in iterative design and optimization. This paper develops a machine learning based surrogate model capable of predicting steady state and transient characteristics of SOI MOSFETs using device geometry, doping profiles, and bias conditions as inputs. A deep neural network (DNN) with hyperbolic tangent activation functions are trained with randomized initialization and early stopping to ensure stable convergence and smooth derivatives compatible with SPICE simulation. The proposed model achieves an average prediction error of 0.81% and a maximum deviation of 4.3%, while providing nearly 600 × speed-up relative to TCAD simulations. These results demonstrate that machine learning (ML) assisted modeling is an accurate and computationally efficient alternative for SOI MOSFET characterization and compact modeling.