| Summary: | Machine Learning (ML) is one of the fastest growing technologies in recent years, having the ability to add value to workflows and to existing technologies. An aspect of ML that is present in many enterprise applications is the detection of anomalies. This project aims to create a system for call anomaly detection in a contact center context, more precisely detection of audio cuts. For this, it was researched the best features and models for the solution, by understanding the original data set; re-labeling it to improve the data representation; extracting relevant features, to distinguish the classes; and selecting the most relevant ones to the system. The models used to create the system were the Support Vector Classifier (SVC) and the Random Forest Classifier, the last one having shown the best performance. A clustering-based approach was also performed on the class that represented the calls with worse audio quality, through the implementation of the K-means algorithm, revealing the possible stratification of two different types of calls within this class. The results showed that the Random Forest was the best performing model, so it was used in the final solution. This solution was inserted into a web app for integration in a business context allowing to improve the Quality of Service (QoS) in contact centers.
|
|---|