In the dynamic world of Mobile Network Operators (MNOs), meeting Service Level Agreements (SLAs) is crucial. To avoid breaches, continuous monitoring of network performance is necessary, proactively addressing and preventing issues before they reach the customer's doorstep. Therefore, Root Cause Analysis (RCA) plays a key role in swiftly resolving failures and determining effective solutions.
However, with the rapid advancement of technology, traditional RCA approaches are gradually being overshadowed. As networks embrace flexibility and complexity, engineers are now turning to cutting-edge Machine Learning (ML)-based solutions for fault detection and diagnosis. But here's the catch - these ML systems often lack explainability, working like mysterious black boxes and making it hard for network engineers to grasp the reasons behind faults.
In this blog, we explore the challenges faced by modern Smart RCA (SRCA) systems, uncovering the balance between automation and explainability that holds the key to seamless services and customer satisfaction. Moreover, we will present the innovative work developed by Celfinet - A Cyient Company in this area.
Demystifying ML Models: Bridging the Explainability Gap
In the dynamic telecommunications landscape, unraveling the intricacies of ML models stands as a pivotal challenge. Within the realm of Celfinet - A Cyient Company, we not only acknowledge the significance of this obstacle, but also recognize its critical role in shaping reliable solutions. This challenge is underlined by the often-complex decision-making of Artificial Intelligence (AI), which adds complexity and can obscure transparency and trust.
In light of these considerations, we emphasize the importance of eXplainable Artificial Intelligence (XAI) in tackling this challenge. This strategic approach empowers us to provide our clients with actionable insights into their network's performance. By understanding the rationale behind model predictions, stakeholders can make informed decisions, enhance network efficiency, and proactively address potential issues, as illustrated in Figure 1.
Figure 1- Implementation Framework for a Smart RCA System Based on XAI.
In seeking optimal solutions for explainability, two distinct approaches stand out. Black-box models excel in predictive capabilities but need post-hoc techniques, such as SHapley Additive exPlanations (SHAP) and Local Interpretable Model-agnostic Explanations (LIME), to explain their decision-making processes. In contrast, glass-box models, like Explainable Boosting Machine (EBM) and Logistic Regression (LR), prioritize transparent decision-making and offer clearer insights into the model's behavior, although they generally have lower predictive performance. Therefore, choosing between these options involves a trade-off between explainability and predictive capability based on specific requirements and the importance of transparency in a given context. Therefore, the field of XAI remains in a constant state of evolution, garnering increased attention and traction as it continues to shape the ever-evolving technological landscape. This journey aligns seamlessly with Cyient's commitment to fostering a future in the telecommunications industry that thrives on the transformative potential of AI. By combining advanced methods, our objective is to develop easily understandable SRCA solutions for our clients, bolstering network reliability, while ensuring a seamless experience for end-users.
Unmasking Network Issues: Celfinet’s Journey into Explainable AI-Powered Fault Diagnosis Techniques
With a focus on addressing explainability challenges and driving telecom transformation via AI, our work has been yielding noteworthy results. A significant advancement is presented in our scientific publication titled 'Root Cause Analysis of Low Throughput Situations Using Boosting Algorithms and TreeShap Analysis’. This paper delves into the complex domain of mobile networks, where identifying the underlying factors causing low-performance scenarios stands as a prominent concern. In response to this challenge, we employ a synergistic blend of black-box supervised learning techniques and the SHAP method. By harnessing the capabilities of classification models, this paper adeptly forecasts and localizes instances of low User Downlink Average Throughput, thus proactively addressing network performance bottlenecks. This method automatically finds crucial Key Performance Indicators (KPIs) that affect the model's predictions, aiding in pinpointing significant performance factors behind low throughput occurrences.
Building on this research, our publication titled 'Explainable Fault Analysis in Mobile Networks: A SHAP-based Supervised Clustering Approach' represents an advancement in fault diagnosis. This new framework not only successfully detects anomalies within the User Downlink Average Throughput KPI but also automatically categorizes them into distinct fault clusters. These clusters inherently encompass diverse root causes, including radio conditions, low network usage in specific user groups, low network capacity, and mobility issues.
Lastly, our latest publication titled 'Unraveling the Root Causes of Faults in Mobile Communications: A Comparative Analysis of Different Model Explainability Techniques' compares the performance of the methodology developed in the previous paper, utilizing both black-box models with post-hoc techniques and glass-box models. The study concludes that black-box models paired with the SHAP method showed superior performance in fault detection and diagnosis while maintaining explainability.
The key findings of the most recent SRCA algorithm, developed by Celfinet, are illustrated in Figure 2 and extensively discussed in the aforementioned papers.
Figure 2 - Methodology and results obtained through Celfinet's SRCA model.
This research underscores our commitment to advancing smart solutions at Celfinet – A Cyient company. Through our research department, we are dedicated to pushing the boundaries of innovation. This commitment empowers mobile network operations to leverage performance enhancements, further solidifying our contribution to the landscape of Smart Network Operations Centers (NOCs).
The Path Ahead: Unfolding the Future
With growing reliance on mobile networks and increased complexities, the need for innovative solutions to control and improve network performance is crucial. In this evolving landscape, AI takes centre stage as a revolutionary instrument, endowing network operators with an arsenal of cutting-edge automation and robust capabilities.
At Celfinet - A Cyient Company, we are constantly vigilant to the advancements in the field of AI, ensuring their integration into our solutions. As the digital revolution gains momentum, our unwavering mission involves not only driving a paradigm shift in the smart operations landscape, but also innovating solutions that go beyond traditional limits.
About the Author
Madalena Cilínio is a research engineer at Celfinet - A Cyient Company, where she has been working for the past two years in Celfinet's Research Department. She completed her master's degree in electrical and computer engineering from Instituto Superior Técnico in 2022. Madalena's main focus revolves around smart operations, where her expertise contributes to the development of AI solutions that finely optimize mobile network performance. More recently, she achieved first place in the 32nd edition of the Young Engineer Innovation Award, organized by the Southern Region of the Portuguese Order of Engineers, with her work "Detection and Diagnosis of Faults in Mobile Networks through Supervised Clustering of SHAP Values”.
Let Us Know What You Thought about this Post.
Put your Comment Below.