To address the challenges of low sensitivity and the difficulty in simultaneously optimizing sensitivity and range in comb-type capacitive pressure sensors, this paper proposes a novel beam-membrane structured comb-type capacitive pressure sensor. An optimization approach combining curve fitting with BP (Backpropagation) and NSGA-II (Non-dominated Sorting Genetic Algorithm II) methods is utilized to enhance the sensor′s performance. By introducing anchor points and cantilever beams to the diaphragm, creating a lever amplification structure, and connecting the movable comb fingers to the lever′s output, the displacement of the comb fingers is amplified, improving sensitivity. To handle the high dimensionality and substantial computational demands of the dataset, MATLAB is employed for data fitting and quantitative analysis of the structural and performance parameters. A correlation analysis between geometric parameters (such as anchor points and cantilever beams) and performance metrics identifies key factors influencing sensor performance, allowing for the elimination of redundant variables and reduction of dataset complexity. The dimensionality reduction process decreases the dataset from 14 to 6 dimensions without compromising accuracy, thus enhancing data collection efficiency and reducing computational resource consumption. The reduced dataset is trained using a BP neural network, and the NSGA-II algorithm is applied for co-optimization of sensitivity and range, improving output reliability. The results show that within the 0~50 kPa range, the optimized sensor achieves a sensitivity of 0.106 pF/kPa, a 30.4% improvement, with a non-linearity error of 0.4% F.S. This optimization methodology provides valuable insights for refining complex structures with multiple parameters. The proposed sensor, with its enhanced sensitivity and reduced nonlinearity, offers an innovative perspective for advancing MEMS pressure sensor technology.