Detect Difficult-to-find PC Performance Issues through Big Data Machine Learning

Enhancing Efficiency and Accuracy in PC Performance Monitoring

Background of the Problem

Detecting non-functional issues on Laptop or PCs is time-consuming and prone to human error, since it entails monitoring metrics data continuously and manually. Combining this with simulating real-life scenarios can be challenging.

Objective, Solution and How We Do It:

Our objective is to build an integration of a data gathering tool and machine learning model that can:

Extract and store data from CPU, battery, Wi-Fi, GPU, disk space, memory usage, network performance, and system events from test machines.
Detect anomalies in time-series data through a combination of Recurrent Neural Network (RNN) and rule-based models.
Identify real-issue-posing anomalies through Long Short-Term Memory Network (LSTM) AutoEncoders model.
Provide insights through visualized anomaly reporting

Tech Stack:

Solutions: Artificial Intelligence, In-house Server Architecture, Data Gathering Tool
Services: Big Data, Machine Learning, Data Visualization through Tableau, Data Warehousing

Key Results

Detected difficult-to-find performance issues within a 3-day turnaround time

Established a scalable end-to-end system for anomaly detection and reporting

Assisted SWPA in raising defects or real issues from reported anomalies

Doubled real issue detection accuracy from 29% to 60%

Transform your business with Ubicom

Achieve operational excellence and drive growth with our services and customized solutions.