In this paper, we present a Deep Neural Network (DNN) to predict the current-voltage characteristics for feedback field-effect transistor (FBFET) devices. A Savitzky-Golay filter is used to remove noise from the output curve of the DNN and to make the transition in the predicted IV curve steeper. We fetch important device parameters like voltage threshold (Vth) and ON-current (Ion) with low deviation from actual values. We also compare the performance of DNN model with various regression models like Random Forest Regression (RFR) and Support Vector Regression (SVR) to demonstrate that our proposed model is superior to these methods. We train our model with a small data set of TCAD simulations and obtain high accuracy. Our work shows that a DNN along with a Savitzky-Golay filter can be used to understand sharp-switching device characteristics, even without a complete knowledge of the underlying physics. This can prove to be crucial for the design optimization of novel devices.