Cellular Traffic Prediction Using Low-Complexity Decision Trees
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing network management systems face challenges in accurately predicting cellular traffic due to the complexity of modern networks, with current methods like autoregressive statistical prediction systems being highly inaccurate and machine learning-based methods requiring significant computational resources and time, making them impractical for embedded processing units.
Innovation Solution
A low complexity decision tree-based traffic predictor model that is trained using hypothesis testing to select statistically significant features, allowing for fast and accurate traffic predictions on embedded processing units with reduced computational requirements.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If autoregressive statistical prediction systems or machine learning models (LSTM, CNN) are used for traffic forecasting, then prediction accuracy is improved, but computational complexity and resource consumption increase significantly
Solution Approach 1:
The patent extracts only the most statistically significant features from the available network data using hypothesis testing, rather than using all available features or complex model architectures. This selective extraction maintains prediction accuracy while significantly reducing computational complexity and resource requirements for embedded processing units.
Solution Approach 2:
The patent replaces expensive, resource-intensive machine learning models (LSTM, CNN) with a simpler, more efficient decision tree-based model that consumes fewer computational resources. This substitution achieves comparable prediction accuracy while being suitable for deployment on embedded processing units with limited capabilities.
2Measurement precision
If complex machine learning models are deployed at base stations, then prediction capability is improved, but resource consumption and operational expenses increase
Solution Approach 1:
The patent extracts only the essential statistical features that pass hypothesis testing significance thresholds, eliminating redundant features that would consume unnecessary computational resources. This selective feature extraction reduces energy consumption while maintaining the core prediction capability needed for network management.
Solution Approach 2:
The patent substitutes resource-heavy machine learning models with a lightweight decision tree model that requires minimal computational resources to execute. This model can be deployed on embedded processing units at base stations, reducing operational expenses and energy consumption while providing sufficient prediction capability for dynamic resource allocation.
3Measurement precision
If comprehensive feature sets are used for training, then model accuracy is improved, but training time and computational resources increase
Solution Approach 1:
The patent uses hypothesis testing to identify and extract only the statistically significant features from the comprehensive feature set before training the model. This pre-filtering step reduces the dimensionality of the training data, significantly decreasing training time and computational resource requirements while retaining the features most important for accurate predictions.
Data Source
AI summary
The technology described herein is directed towards accurate low-complexity decision tree-based traffic predictor models, such as decision tree regressor models for use by base stations. Each model is rapidly retrained per base station using data relevant to the base station. To improve accuracy, statistically significant feature data is selected by performing hypothesis testing on candidate features to filter out features that cannot satisfy a statistical significance threshold (e.g., p-value). The decision tree regressor model is recursively grown based on the selected features' feature values and their traffic level labels. Predicted traffic level data is determined by traversing the trained decision tree to reach a leaf node associated with the prediction data. Resource allocation can be based on the prediction. In addition to time-series training data, spatial training data can be used. Real time traffic monitoring by a radio unit for operating in an autonomous management mode is also facilitated.