<p>This paper introduces a novel approach to channel estimation in reconfigurable intelligent surface (RIS)-assisted millimeter-wave (mmWave) systems, leveraging machine learning to enhance traditional compressive sensing methods. In mmWave communication, the rapidly changing channel conditions and inherent signal sparsity present significant challenges for accurate channel estimation. To address these challenges, we propose a hybrid model that combines deep learning with compressive sensing, enabling adaptive and efficient channel estimation. The deep learning component applies a convolutional neural network (CNN) to interpret and predict complex channel dynamics based on previous data, and the compressive sensing approach takes advantage of the sparsity of mmWave channels to minimize pilot overhead. By incorporating these approaches, our system dynamically adjusts to diverse channel conditions, improving channel estimation accuracy and minimizing computing complexity. Extensive simulations demonstrate that the proposed method surpasses conventional estimating methods, such as orthogonal matching pursuit (OMP), in terms of normalized mean square error (NMSE) and adaptability to non-ideal conditions.</p>