Abnormal price movements in stock markets can signal manipulation, liquidity shocks, or emerging systemic risk, yet conventional statistical detectors and existing LSTM-Autoencoder systems lose accuracy under volatile conditions and rarely provide real-time alerting or rich reporting. This study develops a stock-price anomaly detection model using a hybrid Deep Neural Network (DNN) and LSTM-Autoencoder, enhanced with volatility-aware and technical features. We cleaned daily Open, High, Low, Close, and Volume data from Yahoo Finance, Min-Max normalised it, enriched it with technical, volatility, and rollingstatistical features, and converted it into time-series sequences. The model learns normal market behaviour without labelled anomalies and flags observations whose reconstruction error exceeds a threshold. It was implemented in Python with TensorFlow/Keras, served through a FastAPI framework with real-time alerting, and backed by a PostgreSQL database and an interactive dashboard with downloadable reports. Compared with the existing LSTM-Autoencoder system, the proposed model achieved an accuracy of 97.60% versus 94.80%, precision of 96.90% versus 94.10%, recall of 95.80% versus 93.20%, an F1-score of 96.35% versus 93.65%, and a ROC-AUC of 98.40% versus 95.70%. False positive and false negative rates fell from 5.10% to 2.80% and from 6.80% to 4.20%, and average detection time fell from 1.84 s to 1.12 s. The findings indicate that combining DNN layers with LSTM-Autoencoder learning and volatility-aware features improves both the accuracy and the speed of stock anomaly detection, supporting its use for market surveillance and risk management.

O. G. Mbaja, V. I. E. Anireh, O. E. Taylor, "Hybrid DNN-LSTM Autoencoder Model for Anomaly Detection in Stock Prices Using Volatility-Aware and Technical Features" International Journal of Computer Science and Engineering Development, V1(4): Page(142-147) September - October 2026. ISSN: 3139-0862. www.ijcsed.com. Published by Scientific and Academic Research Publishing.