In this work, an adaptive machine learning-based temperature prediction model is proposed to forecast the system temperature precisely. By using the LMS-based weight adjustment method, the proposed temperature prediction model can adapt to the temperature behavior dynamically. Compared with the related works, the proposed temperature prediction model can reduce 37.2% to 62.3% average error and 36.8% to 88.7% maximum error. With the precise information of the temperature prediction, the involved PDTM can control the system temperature properly and helps to improve the system throughput by 9.16% to 38.37% and bring smaller area overhead than the related works by 18.59% to 22.11%.