Temperature changes caused by thermal effects greatly impact the performance of lithium-ion batteries. It is necessary to figure out the source of heat to assist battery thermal management, and to predict the battery temperature in order to warn the abnormal situation. Herein, this work demonstrate a series of data-driven approaches to analyze the battery thermal effects. From the perspective of time series data, we decompose the temperature change during the battery operation to distinguish the reversible heat and the irreversible heat. The strong correlation between reversible heat and charge/discharge current is verified. Meanwhile, it is found that the irreversible heat have a severe effect on the overall temperature in the later period of battery lifespan. Besides, the long short-term memory (LSTM) model is applied to predict the battery temperature changes. Relying on this machine learning method, we can accurately compute the temperature value of the battery at a certain time. Moreover, using the average temperature of each cycle as the training data, the temperature fluctuation of the battery can be efficiently predicted in a long period, which can serve as the battery temperature prognostic.