The mutation serves as a biomarker for cancer diagnosis and prognosis, indicating early-stage cancer. Efforts are underway to develop advanced pre-cancer detection methods and therapeutic strategies to restore TP53 function or counteract its loss. This study evaluates the performance of different deep-learning techniques in mutation detection as a precancer classifier. The evaluation was conducted in two distinct phases. Following data processing and feature selection, the performance of several models—Artificial Neural Network (ANN), Decision Tree (DT), K-Nearest Neighbors (KNN), and Convolutional Neural Network (CNN)—was systematically assessed across accuracy, sensitivity, and specificity metrics for codons 248, 249, and a combined dataset of these codons. Each model's effectiveness was evaluated under two feature conditions, encompassing eight and thirty-three features. The Decision Tree, identified as the optimal performer, was further enhanced by integrating it with the XGBoost deep learning algorithm to maximize performance. Integrating a DT with XGBoost improves accuracy from 93.15% to 96.55%, sensitivity from 94% to 98%, and specificity from 92% to 96%, making it more effective in detecting precancers based on codon mutations. This combined model enhances both true positive and true negative identification. The detection of codon mutations shows promise for early cancer detection.