Rock mechanical parameters are among the most critical factors in petroleum exploration and development. The buried hill reservoirs in the Bozhong area are characterized by well-developed fractures and strong heterogeneity, posing significant challenges for accurate parameter prediction. Traditional prediction models, which rely on empirical formulas or physics-driven methods, suffer from limitations such as insufficient consideration of influencing factors, poor data adaptability, inadequate capture of nonlinear characteristics, and weak generalization capabilities, making them unsuitable for accurately predicting rock mechanical parameters in such complex reservoirs. Machine learning methods, by autonomously uncovering hidden patterns in data, offer a novel, efficient, and high-precision approach to predicting rock mechanical parameters. In this study, core samples from the buried hill reservoirs in the Bozhong area were collected, and uniaxial and triaxial compression tests were conducted to establish a profile of rock mechanical parameters. A fully connected neural network (FCNN) model was developed, utilizing data from two wells for training to predict the rock mechanical parameters of a neighboring well. The results demonstrate excellent predictive performance, with relative errors for the four parameters ranging from only 0.31% to 2.18%. This indicates that the data-driven FCNN model provides a high-precision intelligent solution for predicting rock mechanical parameters under complex geological conditions.