ML Model Selection for Time Series Forecasting
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Developers face challenges in selecting the best machine learning-based models for time series forecasting due to the variety of models available, requiring extensive development effort and computational resources, and needing extensive knowledge of machine learning techniques.
Innovation Solution
A system that generates machine learning-based models for time series forecasting by determining a model metric based on application characteristics, accessing a pool of models, and selecting the best model based on comparison of metric values, with the option to use clustering for metric selection when not provided by the user, and employing a two-dataset training approach to prevent overfitting.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If developers manually select and train machine learning models for time series forecasting, then model accuracy can be optimized, but development time and computational resources increase significantly
Solution Approach 1:
The system pre-trains multiple machine learning models on historical time series data before they are needed for a specific forecasting task. This preliminary action allows the system to quickly select and fine-tune the best model without requiring developers to train models from scratch, significantly reducing development time while maintaining forecasting accuracy.
Solution Approach 2:
The system creates copies of pre-trained models and adapts them to specific time series forecasting tasks rather than training new models from scratch. This copying approach allows rapid deployment of models with high accuracy while reducing the computational resources and time required for model development.
2Measurement precision
If multiple machine learning models are trained and evaluated to select the best model, then forecasting performance improves, but computational resources and development effort increase
Solution Approach 1:
The system performs preliminary training of multiple machine learning models on historical data in advance. This pre-computation allows the system to quickly evaluate and select the best model for a specific task without requiring extensive computational resources at the time of actual forecasting, thus reducing overall computational resource consumption while maintaining high forecasting performance.
Solution Approach 2:
The system segments the model selection process into distinct phases: pre-training multiple models, evaluating them against performance metrics, and selecting the best one for the specific task. This segmentation allows efficient resource utilization by performing computationally intensive tasks only when necessary, reducing overall computational resource consumption.
3Adaptability or versatility
If developers manually build machine learning models, then model customization is possible, but extensive knowledge and expertise are required
Solution Approach 1:
The system automatically selects, trains, and evaluates machine learning models based on the characteristics of the time series data and the forecasting task. This self-service capability eliminates the need for developers to have extensive machine learning expertise, while still allowing for model customization through configuration parameters and data input, thus making the system easy to use for users with varying levels of technical knowledge.
Solution Approach 2:
The system acts as an intermediary between the user and the complex machine learning modeling process. Users simply provide time series data and forecasting requirements, and the system handles the model selection, training, and evaluation automatically. This intermediary approach maintains model customization capabilities while eliminating the need for users to have extensive machine learning expertise.
4Measurement precision
If extensive model evaluation is performed to select the best model, then forecasting accuracy improves, but time and computational resources are wasted
Solution Approach 1:
The system pre-evaluates multiple machine learning models against various performance metrics before they are needed for a specific forecasting task. This preliminary evaluation allows the system to quickly identify and select the most promising models without performing extensive evaluation computations at the time of actual use, thus reducing computational resource waste while maintaining high forecasting accuracy.
Solution Approach 2:
The system performs partial evaluation of models using a subset of historical data and key performance metrics rather than exhaustively evaluating all possible models on all available data. This partial action approach is sufficient to identify the best models for the specific task, reducing computational resource consumption while still achieving high forecasting accuracy.
Data Source
AI summary
A system according determines a machine learning based model for forecasting time series data for a given use case. The system determines a model metric for a specific use case of time series data. The system accesses a pool of machine learning based models including a plurality of machine learning based models machine learning based models based on different machine learning techniques. For each of the plurality of machine learning based models the system performs forecasting using the machine learning based model and determines the value of the model metric for the machine learning based model. The system selects a machine learning based model based on comparison of values of the model metric for machine learning based models. The system uses the selected machine learning based model for forecasting values for the time series data for the application.