Multistage plug and perforation (plug-n-perf) fracturing is commonly used for horizontal well completion in unconventional reservoirs. Uniform distribution of proppant across all clusters in each stage has proved to be challenging with low viscosity slickwater owing to its limited transport capability. Computational fluid dynamics (CFD) has been used to model proppant transport in wellbore to improve perforation and fracturing design for achieving uniform proppant placement. However, traditional CFD modeling of a full-scale stage is computationally expensive, which limits its applicability in the completion design optimization. A new approach was developed in this paper to efficiently predict proppant placement along a multicluster stage based on a machine learning (ML) model trained with extensive CFD modeling results. Its high computational efficiency permits quick sensitivity analyses to optimize perforation and fracturing designs. The new approach was validated against full-stage CFD modeling results as well as post-treatment field diagnostics. Sensitivity analyses show that proppant inertia effect is a key factor affecting proppant placement in heel clusters with higher slurry flow rates, allowing more proppant carried to the toe owing to its higher density in comparison with fluid. Proppant settling allows bottom perforations to accept more proppant than top perforations. This gravitational effect is not negligible near the heel at high flow rates and becomes more dominant near toe clusters where the flow rate is reduced. Near-uniform proppant placement is achievable via perforation design optimization by taking advantage of these two key mechanisms controlling proppant transport in horizontal wellbores. It is demonstrated that in-line perforating designs with all perforations having the same orientation in each cluster or the entire stage, especially with perforations at the bottom or on the side of the wellbore, improve the proppant placement uniformity. However, it is recommended that the optimum perforation design should be identified case by case depending on specific input parameters. The ML-based model developed in this study has overcome some of the limitations from existing models in the literature and is able to provide quick and yet reliable solutions to proppant placement prediction and design optimization.