<p>The high porosity of recrystallized silicon carbide (RSiC) is a key drawback limiting its performance improvement, thereby developing the low-porosity RSiC is still a popular topic until now. In the previous studies, the pursuit of low-porosity RSiC had typically relied on a trial-and-error approach to adjusting various parameters, which suffered from long cycles, low efficiency, and high costs. Taking RSiC as the research object, this study introduces a normal equation-based machine learning workflow in multiple linear regression for predicting its porosity, water absorption, and bulk density and enabling the inverse design of particle gradation from targeted properties. Leveraging the minimal post-sintering densification of RSiC, particle gradation was targeted as the critical factor controlling green density. Using a dataset of 117 samples, an optimal regression model was developed and selected to predict porosity, water absorption, and bulk density for various particle gradation. The model achieved high reliability, with nearly all relative errors below 5% and a mean relative error of merely 2.68%. Furthermore, the inverse prediction could be successfully implemented by the established model, i.e., giving the required particle gradation after inputting desired porosity. The data revealed maximum and mean absolute errors in porosity prediction of 0.5% and 0.27%, respectively, while maximum and mean relative errors were only 3.5% and 1.84%, which were much lower than the reported error margins (The average relative error is 12.73%). This research establishes an alternative pathway to solving the issue of elevated porosity in RSiC.</p>