The reliability and efficiency of large-scale complex network systems, such as those found in data centers, cloud infrastructures, and telecommunications, are essential to modern society in this age of digital revolution. Large volumes of data are continuously generated by these systems, which makes it difficult to monitor, identify, and diagnose problems in an efficient manner. Because of these networks’ immense size and complexity, traditional fault detection techniques frequently fall short, leading to sluggish responses and possibly catastrophic system outages. In order to successfully handle these issues, this work presents a novel method that combines t-Distributed Stochastic Neighbor Embedding (t- SNE) with Fault Tree Analysis (FTA) to remove all anomalies from the dataset and then use Rough Set Theory (RST) to reduce complexity. Using t-SNE to visualize high-dimensional data and clustering, combining RST for data reduction and detecting important features. Our approach improves the precision and effectiveness of fault detection and diagnosis by combining t-SNE for high-dimensional data visualization for unusual values in dataset with FTA for anomaly cause analysis. Significant gains in fault detection rates and diagnosis results shown by experimental validation using a simulated large-scale network dataset, highlighting the potential of this hybrid approach in reliably managing and sustaining robust network operations amid the growing volumes of data.