Fracturing parameter design is critical to the economic viability and effectiveness of reservoir stimulation, making multi-stage hydraulic fracturing in horizontal wells a cornerstone technology for unconventional hydrocarbon development. However, conventional design methods relying on mechanistic models and empirical expertise often fail to provide well-specific customization due to their limited adaptation to geological heterogeneity. Erratic fracturing performance often stems from the limitations of existing data-driven approaches in integrating multi-source data. To address these challenges, this study proposes a data-driven fracturing parameter design strategy based on the fusion of well-logging data. The approach utilizes a hybrid architecture combining a convolutional neural network (CNN), a multi-layer perceptron (MLP), and a comprehensive learning particle swarm optimization (CLPSO) algorithm. Specifically, the Gramian angular field (GAF) is implemented to transform logging data into visual representations, enhancing feature extraction and production prediction accuracy to inversely derive optimal operational parameters. Results demonstrate that the proposed model substantially improves production prediction accuracy and enhances single-well output through parameter optimization. Compared to the AutoGluon ensemble framework, the proposed model achieves a 20% improvement in prediction accuracy. Field applications demonstrate that the model increases single-well shale oil yields by 6.6%–18.1% compared to conventional designs. This study provides a novel framework for optimizing fracturing parameters in unconventional shale oil reservoirs.