Cybersecurity risks, particularly attacks using Trojan horses, represent a significant risk to modern computer systems. We suggest and evaluate improved deep learning methods for Trojan horse attack detection using the Kaggle-hosted Trojan Detection dataset. Deep learning may help cybersecurity identify Trojan horses. These neural network methods find bad trends in massive data. Deep learning models re more accurate and efficient than traditional detection approaches because they adapt to changing threats. Recent research demonstrates that new Trojan horse detection methods are quicker and more accurate. It prevents data breaches and unauthorized access, making it a vital cybersecurity improvement. This work makes use of these techniques. To achieve a high degree of accuracy in identifying malicious software, our technique is based on the classification of convolutional neural networks (CNNs). We preprocess and enrich the dataset to enhance the model’s generalizability. As a result of extensive testing and optimization, Trojan horse attacks and other cyberattacks risk sensitive data and systems. Trojan horses and malicious software masquerading as legitimate, may be catastrophic. CNN-based classification solves this. From Kaggle’s Trojan Detection dataset. Data preparation and enrichment increase feature representation. Models classify threats. This proposed Model achieved an impressive accuracy rate of 98.6% in identifying Trojan horse attacks in cybersecurity applications.