OFDM is a ubiquitous modulation scheme used to achieve high data rate during transmission and reception in broadband internet of things application. But the envelope aberration ofOFDMsignal leads to high peakto-average power ratio (PAPR) which finally results in overall transmitter inefficiency. This article proposes a neural network (NN) based gradient clipping approach at the transmitter and a linear regression model at the receiver to minimize the PAPR in OFDMsystems with reasonable computational complexity. The simulation result shows that when compared to the original OFDM signal, the proposed Neural network-based Regression assisted PAPR Reduction achieves PAPR reduction by 82.9% and 84.1% for 64QAM-OFDM and 16QAM OFDM signal respectivelywithout compromising on the bit error rate (BER) performance.