The traditional method of taking attendance using paper sheets is prone to errors like impersonation, loss, or theft. To solve this issue, automatic attendance systems utilizing identification technology such as barcode badges, electronic tags, touch screens, magnetic stripe cards, and biometrics have been implemented. Biometric technology uses physiological or behavioral characteristics for identification purposes, but traditional biometric systems have limitations such as vulnerability to damage or alteration over time, and variations in occlusions, poses, facial expressions, and illumination can affect face recognition accuracy. Fingerprint identification relies on the distinctiveness of fingerprints and involves comparing two impressions of the friction ridges on human fingers or toes to determine if they belong to the same individual. There are five primary categories of fingerprints: arch, tented arch, left loop, right loop, and whorl. Various algorithms have been developed to recognize fingerprints using minutiaebased matching, which involves identifying key features like ridge ending and bifurcation. Deep learning algorithms, particularly convolutional neural networks, have been successful in improving identification accuracy by extracting features automatically from fingerprint images. In recent times, securing personal data has become increasingly important, and the Convolutional Neural Network (CNN) identification system is recommended for improving accuracy and performance. This paper proposes a fingerprint identification system that combines three models: CNN, Softmax, and Random Forest (RF) classifiers. The conventional system uses K-means and DBSCAN algorithms to separate the foreground and background regions and extracts features using CNNs and dropout approach. The Softmax acts as a recognizer. The proposed algorithm is evaluated on a public database and shows promising results, providing an accurate and efficient biometric identification system.