This repository contains the dataset and supporting files used in the study "Neural Networks for Direct Material Distribution in Frequency Topology Optimization". The dataset was generated through 10,000 automated frequency-based topology optimization simulations performed using the Bi-directional Evolutionary Structural Optimization (BESO) method on a two-dimensional cantilever beam benchmark. The simulations include randomized optimization parameters, loading conditions, mesh discretizations, and geometric constraints represented by circular no-design regions (holes). The repository contains: problem_configs.csv: input parameters and configuration data for each topology optimization simulation, including optimization settings, loading conditions, mesh characteristics, and hole geometry parameters. beso_results.csv: output performance indicators associated with each simulation, including optimization metrics and computational time. beso_topologies.mat: binary topology matrices generated by the BESO algorithm, representing the optimized material distributions. dataset_cantilever_hole_2026_v4.mat: processed dataset used for neural network training, including resized topology representations and associated input/output data. The dataset was created to support research on machine learning-assisted topology optimization, surrogate modeling, engineering design automation, and data-driven structural optimization. In particular, it was used to train and validate a multi-head neural network capable of directly predicting optimized material distributions and performance indicators, subsequently integrated into a Bayesian Optimization framework. The benchmark problem consists of a frequency-constrained topology optimization of a cantilever beam with variable concentrated masses and randomly generated no-design regions. The objective is to minimize structural mass while preserving the first natural frequency. Researchers are encouraged to use this dataset for benchmarking, reproducing the results presented in the associated publication, developing alternative machine learning models, or investigating data-driven approaches for topology optimization and engineering design. If you use this dataset in your research, please cite the associated publication. Keywords: Topology Optimization, BESO, Frequency Optimization, Neural Networks, Bayesian Optimization, Machine Learning, Engineering Design, Structural Optimization, Additive Manufacturing, Data-Driven Design.