📄 Abstract
Anomaly detection plays a critical role in identifying unusual patterns that may indicate faults, fraud, or security threats across various domains such as finance, healthcare, and network systems. This project presents an AI-driven anomaly detection approach using deep autoencoders, a class of unsupervised neural networks designed to learn efficient data representations. The proposed model is trained on normal (non-anomalous) data to learn its underlying structure by encoding inputs into a compressed latent space and reconstructing them with minimal loss. During inference, the model evaluates new data instances by measuring reconstruction error. Since the autoencoder is optimized to reconstruct normal patterns, anomalous data points typically result in significantly higher reconstruction errors, enabling effective identification of deviations. The project explores different architectures of deep autoencoders, including stacked and sparse variants, and evaluates their performance using benchmark datasets. Experimental results demonstrate that the proposed method achieves high accuracy and robustness in detecting anomalies without requiring labeled anomaly data. This makes it particularly suitable for real-world applications where anomalies are rare and difficult to label. The study also highlights the scalability and adaptability of deep autoencoders in handling high-dimensional data, making them a powerful tool for intelligent anomaly detection systems.
🏷️ Keywords
📚 How to Cite:
Kuna Sujana Sri, Koyilada Prasanthi, D. Aruna Padma, B. Jyothi , Sunil B , AI-DRIVEN ANOMALY DETECTION USING DEEP AUTOENCODERS: A PYTHON AND FLASK-BASED RECONSTRUCTION APPROACH FOR INTELLIGENT MONITORING , Volume 12 , Issue 7, July 2026, EPRA International Journal of Multidisciplinary Research (IJMR) , Pages: 683 - 692 ,