US 2026/0197283 A1Published application

Machine learning approaches for provisioning telecommunications services

AssigneeT-Mobile USA, Inc.

InventorsMohammad Mustafa Bari, Lema Kandula

Life of this patent

Estimating the term…

Loading patent…

Abstract

A computer system receives diverse telecommunications network data including experience metrics or satisfaction scores from multiple sources across a network. The raw data contains inherent imbalances where positive adoption cases represent less than a small fraction of the total cases across service categories. The computer system processes this imbalanced data using under-sampling to create balanced datasets, then performs feature engineering to extract features such as voice band size into weighted feature vectors. The engineered features are used to train specialized AI models that generate propensity scores indicating service adoption likelihood for different telecommunications services. The computer system determines lift scores by comparing the propensity-based predictions against baseline response rates to validate accuracy. When lift scores exceed a threshold, indicating better-than-random prediction accuracy, the corresponding services are identified and provisioned for user devices.

Claims20

Dependent claims

A computer system comprising:

at least one hardware processor;
and
at least one non-transitory memory storing instructions, which, when executed by the at least one hardware processor, cause the computer system to:
receive telecommunications network data from multiple data sources associated with a telecommunications network,
wherein the telecommunications network data is associated with at least one user device, and
wherein the telecommunications network data includes data imbalances associated with multiple telecommunications services;
process the telecommunications network data using under-sampling to adjust the data imbalances and generate processed data;
extract data points from the processed data to generate multiple feature vectors, each corresponding to a respective one of the telecommunications services;
generate, using multiple trained artificial intelligence (AI) models, multiple propensity scores for the at least one user device using the feature vectors,
wherein each of the trained AI models provides a respective one of the propensity scores;
determine multiple lift scores for the at least one user device using the propensity scores, each lift score corresponding to a respective one of the telecommunications services;
and
responsive to one or more of the lift scores exceeding a threshold score,
provision one or more of the telecommunications services for the at least one user device,
wherein the one or more of the telecommunications services correspond to the one or more of the lift scores.

12 more elements

Dependencies are inferred from the claim text.