Knowing accurately the characteristics of the tunnel convergence is an essential task in tunneling, especially when the New Austrian Tunneling Method is applied; this allows making any necessary adjustment to the construction methods in order to avoid unwanted situations such as rock collapse, trapping and jamming of boring machine or geological disasters. This study aims at improving a Multivariate Adaptive Regression Spline (MARS) based model that has been proposed previously to predict the tunnel convergence in weak rocks. To this end, Rough Set (RS) theory was implemented to reduce the parameter attributes. The ground conditions were analyzed and the redundant input parameters which can be eliminated were identified objectively by determining the reduct sets. Field data were compiled from tunnel construction projects in Hunan province (China), which were used as case study. The input parameters included the class index of the surrounding rock mass, angle of internal friction, cohesion, Young's modulus, rock density, tunnel overburden, distance between the monitoring station and the tunnel heading face and the elapsed monitoring time. The performance of the models was evaluated by comparing the measured data to the predicted convergence values using several performance indices, namely the variance account for (VAF), root mean square error (RMSE), relative root mean square error mean absolute percentage error (RRMSE) and the coefficient of determination (R2). The results showed that all the input parameters as were identified as core (intersection of all reducts) except the Young's modulus and rock density which were consequently removed from the model. The computed performance indices before and after attribute reduction, were: 94.26 and 97.02; 0.42 and 0.28; 0.18 and 0.11; 0.95 and 0.98 for VAF, RMSE, RRMSE, and R2 respectively, indicating noticeable improvement. It is concluded that MARS along with RS can constitute a reliable alternative to existing approaches in dealing with nonlinear geo-engineering problem such as the tunnel convergence.