Glassbeam has built a predictive analytics app built for telecommunications operators and providers. This app aims to leverage benefits from our machine learning models. It is designed to automatically turn raw data from networking equipment into actionable operational intelligence. Whether it is solving a support escalation or looking for opportunities to price the services effectively, this app has it. In case you missed it, here is a recent PRESS RELEASE on the QoE app launch.
What’s in the Sauce: The True Picture of the QoE Solution?
There are two parts to the problem: 1. The data is quite complex. (This explains why there are few out-of-box solutions out there and nobody confidently guarantees a ROI from their solutions.) 2. The relevancy and timeliness of action/response define the varying degree of Quality of Experience (QoE).
We also know that the other challenges telecom operators’ have range from their ability to detect negative quality metrics to having a great deal of flexibility in plotting trends on reactive measures. In our discussions with prospects, a point that stood out is that real-time diagnostics solutions for case resolution teams are passé. They believe their best bet is having the ability to detect network performance issues before they actually occur.
Therefore, the tool we created really has one goal: Predictive Support Analytics. We’ve pulled through anomaly detection models trained over petabytes of data to help achieve this goal.
Network Diagnostic Resources Need a Trump Card
For this post, we’ll concentrate on a single problem area: Access Points (APs).
APs include important metrics such as channel utilization, loss, retries, associations, neighbors, and throughput. Our goal is to use these metrics to predict the Quality of Experience (QOE) at both the AP and site level. One important assumption is that the AP metrics collected today have sufficient predictive power. We used machine learning techniques to build this capability.
In the first phase, we built a simple model that predicts whether an AP provides good or bad QoE. We experimented with unsupervised learning techniques but required no labeled data or user input. This was faster to build ─ we rigged up a quick dashboard that illustrated QoE parameters on the fly.
In the second phase, we built supervised learning techniques that required a lot of label data. It was obvious that the accuracy of the model depended on the amount of labeled data while the accuracy increased with the amount of data.
Once we trained the model using our unsupervised learning algorithm, we saved the model parameters in a configuration file. Today, any data-crunching app like LogiXML can then use these model parameters to score new data.
Big Brown Bag of Goodies: QoE on Streaming Data
In the final push to elevate the model’s strength, we put together the capabilities to collect live AP data in a streaming fashion and score them. A dashboard continuously updates the current status of a site using aggregated QoE for a site. The QoE for a site gets calculated using our algorithm using the QoE for each AP at that site. This will allow telecom providers and their customers to automatically detect problematic APs and proactively takes steps to fix problems.
The model boasts uses multi-variate Gaussian distribution techniques. By using complex techniques, we’re able to collect a large amount of labeled QoE data on Access Points, train a classification model, and cross-validate the models to refine the model parameters. With the model examining hundreds of petabytes of network equipment data, the accuracy is the best in the industry.
Taking the Sting out the Big Data Panic
Our QoE solution empowers the data recipients as data experts. The QoE dashboard collects and processes AP data in real time. With a nifty dashboard, users can drill through APs and detect the influence of QoE on a specific site. Do get in touch with us at SUPPORT@GLASSBEAM.COM, we’d be happy to share our experiences in greater detail.