Point cloud-based deep neural networks (PC-DNNs) has seen growing interest in the construction domain due to their remarkable ability to enhance Building Information Modeling (BIM)-related tasks. Among these tasks, Industry Foundation Classes (IFC) object classification using PC-DNNs has become an active research topic. This focus aims to mitigate classification discrepancies that occur during the interoperability of BIM tools for information exchange. However, existing studies have not fully investigated the potential of the PC-DNN models for IFC object classification. This limitation is due to the reliance on a limited number of PC-DNN models trained on small, private datasets that are not openly accessible. To address this knowledge gap, this study evaluates diverse state-of-the-art PC-DNN models for IFC object classification. Our study provides a comprehensive analysis of how different PC-DNN components and loss functions affect IFC classification, utilizing two public IFC datasets: IFCNet and BIMGEOM. Experimental results offer a detailed comparison across metrics such as accuracy, learning progression, computation time, and model parameters.