Ensemble Learning for Time-Series Forecasting
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Time series forecasting is often inaccurate and computationally expensive due to the temporal relationships in data, limiting the adoption and effectiveness of existing statistical and machine learning techniques.
Innovation Solution
The use of ensemble machine learning models, specifically the ELITE (Ensemble Learning for Improved Time Series Estimation) method, which combines forecasts from multiple base learners while preserving temporal ordering, and employs parallel training across distributed computing nodes to enhance accuracy and reduce training time.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional statistical and machine learning techniques are used for time series forecasting, then forecasting capability is provided, but accuracy is poor and computational cost is high
Solution Approach 1:
The patent segments the time series data into multiple non-overlapping folds while preserving temporal ordering. Each fold is used for training and validation in a systematic sequence, allowing the model to learn temporal patterns without requiring complex computational resources for data shuffling or randomization.
Solution Approach 2:
The patent combines multiple base learner models (including traditional statistical methods and machine learning techniques) into an ensemble forecasting system. By merging the predictions of multiple models that each capture different temporal patterns, the system achieves higher forecasting accuracy while distributing the computational workload across simpler individual models rather than relying on a single complex model.
2Productivity
If traditional machine learning models are trained on time series data, then forecasting is produced, but training time is excessive
Solution Approach 1:
The patent performs preliminary segmentation of the time series data into ordered folds before training begins. This pre-organization of data allows for efficient iterative training where each fold is used systematically, eliminating the need for repeated random shuffling and reducing overall training time while maintaining temporal relationships in the data.
Solution Approach 2:
The patent trains multiple base learner models on different subsets of the data (folds) rather than training one complex model on the entire dataset. This partial training approach across multiple simpler models reduces the computational burden on each individual training process while collectively achieving superior forecasting performance through ensemble aggregation.
3Measurement precision
If temporal relationships in time series data are preserved, then forecasting accuracy improves, but computational expense increases
Solution Approach 1:
The patent segments time series data into ordered folds that preserve temporal relationships without requiring computationally intensive methods. By maintaining the chronological sequence in the segmentation process, the system captures temporal patterns efficiently using simple data organization rather than complex algorithms.
Solution Approach 2:
The time series data structure itself provides the temporal relationship information through its inherent chronological ordering. The patent leverages this self-service property of time series data by using the natural temporal sequence in fold segmentation, eliminating the need for additional computational mechanisms to enforce or detect temporal relationships.
Data Source
AI summary
Methods and systems for time series forecasting using ensemble machine learning are disclosed. A computer system (distributed or otherwise) can instantiate, train, and use a plurality of machine learning models to generate time series forecasts. These can include both different types of machine learning models, as well as similar machine learning models that have different configurations. Embodiments of the present disclosure can use a novel modification of k-folds cross validation techniques that preserves the order of temporal data. Time series data can be partitioned into segments and folds and used to train and test the plurality of machine learning models. Forecasts produced by the trained machine learning models, along with historical time series data (or “actuals”) can be used to train an ensemble machine learning model to produce an ensemble forecast based on forecasts generated by the trained machine learning models.