<p>The durability of cementitious materials in aggressive environments, such as those exposed to chloride ions, has been a subject of growing interest in civil engineering, particularly due to the risk of mechanical performance loss. In this context, the present study investigated the compressive strength of mortars with varying silica fume contents exposed to chloride attack addressing the analysis in two stages: (i) experimental tests; and (ii) predictive modeling using artificial neural networks (ANNs). In the first stage, specimens were cast with 0%, 10%, 20%, and 30% silica fume, subjected to two curing conditions (hydrated lime and sodium chloride solutions), and evaluated at three different ages. Results from the first stage showed that the incorporation of silica fume significantly improves mechanical strength, especially in the long term for the 30% composition. Subsequently, the second stage of the research consists of expanding the experimental dataset by generating synthetic data through noise injection, enabling the training of neural networks with different configurations. Architectures trained with 120, 600, and 1200 synthetic samples were tested, varying the number of neurons per hidden layer. Networks trained with larger datasets and between 7 and 12 neurons per hidden layer achieved the lowest prediction errors, with the best validation performance obtained from the architecture with 11 neurons per layer. Finally, the selected models were tested using actual experimental data and presented a mean absolute error below 0.88&#xa0;MPa and strong agreement between predicted and observed values, validating the ANNs’ generalization capability. The results from the second stage of the research confirm the potential of combining synthetic data and artificial intelligence to accurately and efficiently predict the mechanical properties of cementitious materials, even in contexts with limited sample availability.</p>