The point cloud part segmentation task consists of segmenting an object, represented by a point cloud, into its constituent parts, such as a chair that is segmented into seat, backrest, and legs. The most recent computational strategies use Artificial Neural Networks to perform this task, but the architectures used are developed generically and therefore do not consider the specific patterns of each category of objects. Thus, this work proposes to analyze the contribution of building specific architectures based on the optimization of hyperparameters of the PointNet architecture, which is well established in the literature. The dataset used was the PartNet, and four case studies were employed. In addition, we also studied the impact of point cloud size on th is segmentation task, performing the optimization process in each category studied in three different point cloud sizes: 512, 1,024, and 2,048. From the results obtained, an average improvement of 2% in the test accuracy metric was achieved in the Table -1, Chair-1, and Lamp-1 categories and 6% in the StorageFurniture -1 category. The impact of point cloud size was low, and statistically significant improvements were observed in Table -1, Chair -1, and StorageFurniture -1 categories. Thus, hyperparameter optimization proved to be consistent, achieving satisfactory results.