Fingerprint recognition remains crucial in modern security systems; however, challenges persist with low-quality and varied fingerprint images. This study addresses these issues by proposing an Ensemble Convolutional Neural Network-Support Vector Machine (CNN-SVM) model to enhance fingerprint recognition accuracy. Our study investigates the performance of fingerprint recognition models, focusing on preprocessing techniques and ensemble methods to mitigate these challenges. We conducted six experiments to evaluate CNN, SVM, and their ensemble combinations, both with and without preprocessing. Preprocessing steps included data augmentation, normalization, and dimensionality reduction. These experiments were performed using benchmark datasets such as the Sokoto Coventry Fingerprint Dataset (SOCOFing), leveraging a Python programming environment and Google Colab for simulation and analysis. The results demonstrate that the ensemble CNN-SVM model with preprocessing outperforms other configurations, achieving the highest accuracy of 99.73%, precision of 99.79%, recall of 99.73%, and F1 score of 99.74%. This underscores the significant impact of preprocessing and the effectiveness of the ensemble approach in fingerprint recognition tasks. The findings suggest that integrating CNN and SVM models with comprehensive preprocessing steps offers a superior method for addressing fingerprint recognition challenges, contributing to advancements in biometric security systems. Keywords: Fingerprint Recognition, Ensemble CNN-SVM, Data Augmentation, Normalization, Dimensionality Reduction.