This work proposes a machine learning model to predict initial Design Rule Violations (DRVs), using only the information available during the placement stage. The experimental results show good performance for the proposed learning architecture, with 0.93 recall and 0.97 precision, resulting in an F-score of 0.95. Additionally, we present a data augmentation strategy that can be helpful in other machine learning approaches applied in the placement and routing steps of the physical design.