The paper introduces GraphCNNpred, a novel deep learning framework that integrates graph neural networks (GNNs) with convolutional neural networks (CNNs) for predicting stock market index trends. The model leverages a diverse dataset of 82 daily features across five major indices (S&P 500, NASDAQ, DJI, NYSE, RUSSELL). A key innovation is the construction of a correlation-based graph where features are nodes and edges are formed when pairwise correlation exceeds 0.7. This graph enables message passing among correlated features during daily feature extraction, replacing the simple linear transformation used in prior CNNpred models. For durational feature extraction, the model employs CNN layers with 3×1 filters to capture temporal patterns over consecutive days. The framework offers flexible configurations (e.g., GAT-CNN, CNN-GAT, CNN-GAT-CNN) to balance complexity and performance. Experiments show that GraphCNNpred significantly outperforms baseline algorithms (2D-CNNpred, 3D-CNNpred, pure GAT, pure GCN) by 4% to 15% in F-measure across all indices. Additionally, a trading simulation using three-class labeling (up, down, unchanged) achieves Sharpe ratios exceeding 3 and superior CEQ returns compared to always-long and baseline strategies. The results demonstrate the effectiveness of combining GNNs with CNNs for stock prediction, highlighting the importance of capturing both feature correlations and temporal dynamics.