The precise understanding of the nuclear dependence of the neutrino-nucleus cross section is very important in its own right, and in turn helps with the accurate measurement of the neutrino oscillation parameters. We study these issues by examining the event kinematics and cross section ratios between different passive targets at MINERvA. There is an ongoing study of the A dependent nuclear effects in MINERvA; this measurement hinges on identification of the target nucleus, thereby demanding accurate reconstruction of the event vertex. Vertex reconstruction is usually done with tracking-based algorithms. However, the performance of this method suffers when there are tracks created by secondary interactions or decays, or when significant shower activity occurs near the vertex region. Here, we present an alternative vertex finding method in the nuclear target region using machine learning. We use an application of deep learning based on convolutional neural networks (CNNs). Additionally, we explore the use of Deep Adversarial Neural Networks (DANNs) in minimizing possible biases coming from the use of simulated data for the algorithm's training sets.