Intelligent Transportation Systems (ITS) increasingly depend on reliable vehicular links, especially vehicle-to-everything (V2X) technologies and their direct Sidelink vehicle-to-vehicle (V2V) communications. A dynamic selection of the modulation and coding scheme (MCS) is vital to guarantee robust data delivery under the rapidly changing conditions of vehicular networks. In this paper, we present a deep learning framework to forecast optimal MCS settings and assess Sidelink performance, addressing the shortcomings of earlier works that relied mainly on static MCS assignments or ignored real-time adaptation. We implement and compare convolutional neural networks (CNNs), long-short-term memory (LSTM) networks, and gated recurrent units (GRUs), training them on datasets that capture a broad spectrum of channel scenarios and MCS configurations. The results demonstrate that our models can accurately predict communication quality, enabling safer and more efficient ITS deployments.