The rapid proliferation of web applications and internet-based services has significantly heightened the need for robust authentication and intrusion-prevention mechanisms. Among the most critical cybersecurity challenges are unauthorized login attempts and credential-based attacks, including brute force, dictionary, and credential stuffing techniques. Traditional security controls — password policies, CAPTCHA, and signature-based intrusion detection systems — have proven insufficient against these evolving threats. This paper proposes and evaluates a Web-Based Intrusion Detection System (IDS) that leverages deep learning to monitor, analyse, and detect suspicious login activities in real time. The system captures multidimensional authentication features — username, IP address, login frequency, geolocation, device fingerprint, and temporal patterns — and processes them through Artificial Neural Network (ANN) and Long Short-Term Memory (LSTM) models. Experimental results demonstrate that the deep learning-based IDS achieves superior detection accuracy, low false-positive rates, and sub-second classification latency, outperforming conventional rule-based approaches. The proposed system is designed as a modular, scalable architecture comprising frontend, backend, database, and AI inference components, making it applicable to banking, e-commerce, and enterprise environments