Aggregated Machine Learning Model Service for Forecasting
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Business analysts and subject matter experts face challenges in utilizing machine learning models due to the requirement for expertise in statistics and AI, making it difficult to leverage large datasets for improved predictions and decisions.
Innovation Solution
A machine learning service generates an aggregated machine learning model by combining multiple candidate models, each trained with different algorithms, and assigns weights based on error ratios to select the best performing model for prediction, allowing for efficient use of multiple models instead of a single best one.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If multiple machine learning models are trained and aggregated to improve prediction accuracy, then forecasting performance is improved, but the complexity of the system and time required for training increases
Solution Approach 1:
The patent combines multiple machine learning models into a single aggregated model that integrates their predictions. The system trains multiple candidate models with different algorithms and architectures, then merges their outputs through weighted averaging where weights are determined by validation performance. This merging approach improves forecasting accuracy while managing complexity through automated model selection and combination strategies.
Solution Approach 2:
The aggregated machine learning model serves multiple functions simultaneously: it performs prediction, automatically selects optimal model combinations, determines appropriate weighting schemes, and validates performance across different datasets. This multi-functionality consolidates what would otherwise require separate systems into a single universal modeling framework.
2Measurement precision
If multiple machine learning models are trained to improve prediction accuracy, then forecasting performance is improved, but the time required for training increases
Solution Approach 1:
The system performs preliminary training of multiple candidate models on validation datasets before final deployment. By pre-training and evaluating multiple models during the development phase, the system identifies the best-performing models and their optimal weighting combinations in advance, reducing the time required for final model selection and deployment.
Solution Approach 2:
The patent trains more models than strictly necessary (excessive action) to ensure optimal performance is achieved. Multiple candidate models are trained beyond what a single model would provide, with the understanding that not all will be used in the final aggregation. This approach ensures that the best combination is found while managing computational resources through efficient selection processes.
3Measurement precision
If expertise in statistics and AI is required to develop machine learning models, then model accuracy can be improved, but the ease of operation decreases for business analysts
Solution Approach 1:
The system performs self-service by automatically selecting, training, and aggregating multiple machine learning models without requiring user expertise in statistics or AI. The automated framework handles model selection, hyperparameter tuning, performance validation, and weight optimization, allowing business analysts to access sophisticated modeling capabilities through simple interfaces without needing specialized technical knowledge.
Solution Approach 2:
The aggregated modeling system acts as an intermediary between business analysts and complex machine learning algorithms. It translates business requirements into multiple candidate models, automatically manages the technical complexity of model training and selection, and delivers accurate predictions to users who lack deep technical expertise in machine learning.
Data Source
AI summary
Techniques for creating an aggregated machine learning (ML) model from a plurality of candidate ML models are described. According to some embodiments, a machine learning service generates an aggregated machine learning model from a first machine learning model and a second machine learning model, selects the first machine learning model, the second machine learning model, or the aggregated machine learning model for usage, and performs an inference with the selected machine learning model. Additionally, a user may select the candidate models to be trained.