Selecting the most suitable metaheuristic for a given problem instance remains a fundamental challenge in combinatorial optimization. In the context of the Traveling Salesman Problem (TSP), the performance of metaheuristic algorithms varies considerably from one instance to another due to differences in structural and statistical properties, making the selection process highly complex. To address this issue, this study proposes a one-dimensional Convolutional Neural Network (1D CNN) framework for automatic algorithm selection based on meta-learning. Unlike traditional approaches that rely on handcrafted features and static classifiers, the proposed model learns directly from numerical meta-features that characterize TSP instances, enabling it to capture hidden correlations between instance structure and solver performance. The architecture integrates multiple convolutional layers for hierarchical feature extraction followed by dense layers for classification, providing a compact and efficient learning model. Experimental results obtained on a large set of benchmark instances demonstrate that the proposed model achieves a high predictive performance, with an overall accuracy of 93% and balanced precision, recall, and F1-score values. These findings confirm the ability of convolutional feature extraction to model complex relationships within the meta-feature space and to support accurate, data-driven algorithm selection.