Additive manufacturing (AM) has been extensively attracted attention in both academia and industry. AM is known as progressively deposition of material onto a substrate by implementing a thermal heat source. Although AM provides significant improvements in terms of reducing production cost and time, generation of residual stresses inside the fabricated part, as the result of cyclic heating and cooling, is inevitable. Finite elements (FE) analysis has been used as a tool to predict the residual stress distribution in AM parts. Machine learning methods e.g. artificial neural networks have shown great potential in the determination of the relationship between dependent variable(s) and its variables. An FE-based machine learning framework has been introduced in this context to create a robust modeling tool in the estimation of induced residual stresses in AM parts. In this approach, the results of FE-based models of different geometric structures (L-wall, and box) are considered to train a neural network and the trained network is used to predict the residual stress distribution of larger components. A heat transfer analysis is performed on the large parts and the obtained temperature history is used to predict the residual stress distribution of large parts. The initial results show the great potential of this approach in the prediction of residual stress distribution in AM parts by reducing the computational time considerably.