The turbulent Prandtl number (Prt)is an important parameter in heat transfer analysis using RANS (Reynolds-averaged Navier-Stokes equations). However, in a flow under a local acceleration and deceleration, Prt is spatially different. Many variables(velocity gradient, pressure gradient, Pr, etc.) affect the spatial variation of Prt. Principal component analysis(PCA) has been widely used to perform parameter reduction, so this method was adopted to make an effective heat transfer analysis. This study intends to analyze two main objectives. First, ANN(Artificial neural network) performed to verify a small number of new factors, which extracted through PCA, effectively represents the characteristics of many existing variables. Second, PCA performed using all variables including Prt to generate new parameters, combining Prt with the existing parameters. As a result, a new model of Prt derived by selecting those with a large influence of Prt showed an accurate prediction.