AutoML Time Series Forecasting Model Selection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current machine learning approaches for time series forecasting face challenges such as the diversity of time series data characteristics, non-stationarity, seasonal patterns, and the need for hyperparameter tuning, making it difficult to find an optimal model for predictions, especially under time constraints and with limited computational resources.
Innovation Solution
A computer-implemented method and system that prepares time series data, extracts features, generates a feature vector, and inputs it into a classifier model to assess the suitability of various machine learning models for analysis, employing automated machine learning (AutoML) to select the best model based on intrinsic characteristics of the data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If multiple machine learning models are tested in a brute force manner to find the optimal model for time series forecasting, then model selection accuracy may improve, but time consumption and computational resources increase significantly
Solution Approach 1:
The system performs preliminary analysis of time series data characteristics (stationarity, seasonality, trends, autocorrelation) before model testing. This preliminary characterization allows the system to pre-filter suitable models and avoid testing inappropriate ones, significantly reducing time consumption while maintaining accurate model selection
Solution Approach 2:
The patent introduces an intermediary classification model that first analyzes data characteristics and predicts which model type is most suitable. This intermediary step acts as a mediator between raw data and the full model testing process, enabling efficient model selection by guiding which models should be tested based on data properties
2Reliability
If multiple machine learning models are tested comprehensively to assess their suitability, then model assessment quality improves, but computational resources required increase
Solution Approach 1:
The patent segments the model assessment process into distinct phases: data characteristic analysis, model suitability prediction via classification, and targeted model testing. This segmentation allows the system to focus computational resources only on promising model-data pairings rather than exhaustively testing all models, reducing overall computational burden while maintaining assessment quality
Solution Approach 2:
The system changes the parameter of model selection from a exhaustive search approach to a characteristic-driven approach. By analyzing data parameters (stationarity, seasonality, etc.) and using these to guide model selection, the system achieves reliable model assessment with reduced computational resources by avoiding testing of unsuitable models
3Device complexity
If a one-fits-all machine learning model approach is used for time series forecasting, then system simplicity is maintained, but forecasting accuracy decreases due to diversity of time series characteristics
Solution Approach 1:
The patent creates a universal model selection framework that can handle diverse time series characteristics through a single integrated system. The classification model is trained to recognize various data patterns (stationary, non-stationary, seasonal, trending) and automatically select appropriate models, providing multi-functional capability that adapts to different data types without requiring multiple specialized systems
Solution Approach 2:
The system dynamically adapts model selection based on the specific characteristics of each time series dataset. Rather than using a static one-fits-all approach, the classification model analyzes data properties and dynamically determines the most suitable model type for each case, enabling the system to maintain simplicity while achieving high accuracy across diverse forecasting scenarios
4Reliability
If machine learning models are tuned with multiple hyperparameters to achieve optimal performance, then model performance improves, but the complexity of model configuration and training increases
Solution Approach 1:
The system performs preliminary hyperparameter optimization for each candidate model based on the data characteristics and model type predictions. By preparing optimal hyperparameter sets in advance for different model-data combinations, the system reduces the complexity of manual configuration while ensuring each model is tuned for optimal performance on its intended dataset type
Data Source
AI summary
Systems and methods for using machine learning for time series forecasting are disclosed. According to certain aspects, a set of time series data may be prepared and a plurality of features extracted therefrom. A feature vector based on the plurality of features may be generated and input into a classifier model to assess how well each of a plurality of available machine learning models is equipped to analyze the set of time series data and output a time series forecast. In embodiments, a stacking machine learning model may improve the time series forecast by accounting for multiple machine learning models as well as a set of covariates.