AUTONOMOUS SELF-HEALING CLOUD INFRASTRUCTURE USING REAL-TIME TELEMETRY AND PREDICTIVE REMEDIATION

Modern cloud-native architectures face significant challenges with traditional reactive
monitoring,
which often leads to prolonged downtime and alert fatigue. This paper presents a self-healing
framework designed to autonomously detect and remediate system faults in real-time. By
utilizing
real-time telemetry—including CPU utilization, memory pressure, and request latency—the
proposed
system identifies behavioral anomalies through an unsupervised Autoencoder Neural
Network
approach.
The system establishes a dynamic behavioral baseline and detects deviations through
elevated
reconstruction error, enabling it to distinguish between legitimate traffic surges and malicious
anomalies such as memory leaks. Upon detection, a closed-loop automation engine
executes targeted
remediation, such as automated container restarts and rate-limiting, without any human
intervention.
Experimental results, visualized via Prometheus and Grafana, demonstrate that this
autonomous
framework can successfully mitigate slow-burn failures like memory leaks and sudden traffic
spikes.
The system achieved a Mean Time to Recovery (MTTR) of approximately 3 seconds
Chairs have requested users to
enter individual conflicts. Please
click here to enter your individual
conflicts.
30/04/2026, 03:24 Conference Management Toolkit - Submission Summary
https://cmt3.research.microsoft.com/CISCT2026/Submission/Summary/545 1/2
compared to the
90 seconds required for manual recovery—a reduction of over 96%. The framework
effectively
reduces alert fatigue by contextually evaluating anomalies against learned baselines rather
than rigid
static thresholds.
Keywords: Self-healing, AIOps, Cloud Computing, Anomaly Detection, Prometheus, Grafana,
MTTR, Autonomous Systems, Autoencoder, Containerization