Case Study: a global technology firm improves device prediction speed and reduces manual labeling with Persistent Systems' ML models

Global networking leader improves productivity and reduces risk with advanced ML models

The client, a global technology firm, faced a significant challenge with its Business Critical Services team. The team struggled to proactively manage a network of 4.3 million devices because its existing machine learning application was too slow, taking five days to run failure predictions. This delay, coupled with a complex and inefficient manual process for labeling device reset data, hindered proactive maintenance and created data gaps. Persistent Systems was engaged to address these issues with an advanced ML solution.

Persistent Systems implemented a solution utilizing advanced machine learning models to predict device failures and automate reset labeling. The models categorized device risk levels, enabling targeted interventions, while an ensemble of Seq2Seq models and a custom N-gram Regex model automated the labeling process. This approach reduced the average prediction time from five days to just 16-18 hours and improved model accuracy from 85% to 95%. Additionally, the client achieved a 92% reduction in manual labeling efforts.