<p>This study proposes a new approach to predict and optimize additive manufacturing parameters using a variational autoencoder (VAE) model combined with a clustering-based data stratification method. The data set consists of 1,366 samples from the CIRTECH additive manufacturing Center at HUTECH University, which encompasses 20 features divided into three groups: geometric parameters, printing process parameters, and printing outcomes. The proposed method integrates RF to predict process outcomes (such as time and material consumption) and MLP to optimize input parameters based on these predictions. The results indicate that the prediction model achieved high precision with <i>R</i>² = 0.98 and RMSE = 0.16. Although the reverse model had modest performance (<i>R</i>² = 0.24), the overall system ensured stable and reliable prediction capabilities. This approach helps reduce dependence on manual testing, save materials, and improve production efficiency. The study opens up potential applications to optimize the additive manufacturing process and other manufacturing sectors.</p> Graphical abstract <p></p>