In liquid-based battery cooling systems, particularly in electric vehicles and large-scale energy storage, the boiling process is integral for maintaining optimal operating temperatures and ensuring battery longevity and safety. As batteries generate heat through internal resistance and electrochemical reactions, a dielectric liquid coolant is circulated to absorb this heat. Upon reaching its boiling point, the coolant undergoes a phase change from liquid to vapor, efficiently absorbing and removing heat due to the high latent heat of vaporization. The vaporized coolant, carrying the absorbed heat, is then directed through a heat exchanger or condenser where it releases heat and condenses back into liquid, completing the cycle by recirculating to the batteries. In the present study, a combination of machine learning and numerical simulation is used to predict the thermal behavior of the liquid-based thermal management of a lithium-ion battery during a 300s charge-discharge cycle. The cooling system is based on NOVEC 7000 coolant boiling in parallel mini channels. A hybrid code is made to numerically model the boiling process, including how bubbles form and interact with walls in the mini channel. After careful validation of numerical simulation, two distinct machine learning methods are developed in order to predict the thermal behaviour of a cooling system, including flow boiling. Deep forward neural network-boundary condition (DFNN-BC) is used to harvest the pattern of fluid parameters such as pressure, temperature, velocity, and heat transfer of boiling flow. With an 87:13 training ratio, the results of the numerical simulations are used right away to train DFNN-BC. Convolutional neural network (CNN) is then will be used for image processing to keep track of how bubbles form and move during flow boiling. For this goal, bubble formation contours are directly analyzed with an image processing algorithm to find bubble curvature, interface, and the relationship between bubble geometry and flow speed around the bubble using DFNN-BC. CNN will figure out how bubbles will form by looking at 5,000 separate frames of bubbles forming. To reach this goal, the training data set was used with a number of optimization and integration methods, such as ReLu, Maxpool, and feature extraction. An integrator will combine the predictions of DFNN-BC and CNN to examine the thermal behavior of a cooling system over 10,000s.