Domain-Aware Machine Learning Application for Industrial Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current automatic machine learning tools produce sub-optimal models due to the lack of incorporation of domain knowledge specific to the dataset, resulting in less effective predictions.
Innovation Solution
A method that selects a machine learning application specific to an industrial problem, provides standard features for display to a client device, receives a mapping of variables to standard features, applies machine learning pipelines to train models, ranks the models, and selects the best model for prediction.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If automatic machine learning tools are used to create models quickly, then model creation speed is improved, but prediction accuracy deteriorates due to lack of domain knowledge incorporation
Solution Approach 1:
The system segments the machine learning process into two distinct phases: an automated phase that handles standard features and pipeline selection, and a manual phase where domain experts provide domain-specific feature mappings and adjustments. This segmentation allows the system to maintain fast automated model creation while incorporating targeted domain knowledge where it matters most, resolving the contradiction between speed and accuracy.
Solution Approach 2:
The system introduces an intermediary layer between the automated machine learning tools and the domain-specific data. This intermediary consists of domain expert annotations and domain-specific feature mappings that bridge the gap between generic automated algorithms and industry-specific requirements, enabling both rapid development and high accuracy through the mediation of domain knowledge.
2Ease of operation
If generic automatic machine learning tools are used, then ease of operation is improved, but adaptability to specific industrial problems deteriorates
Solution Approach 1:
The system creates a universal platform that can handle multiple industrial problems through a standardized interface. The machine learning application framework provides universal functionality for data processing, model training, and evaluation, while allowing specific industrial domain knowledge to be incorporated through configurable domain-specific features and pipelines. This multi-functionality enables ease of operation across different industries while maintaining adaptability to problem-specific requirements.
Solution Approach 2:
The system dynamically adapts to different industrial problems by allowing domain experts to configure and customize the machine learning pipelines according to their specific needs. The framework provides dynamic flexibility in selecting features, adjusting parameters, and choosing models based on the particular industrial context, while maintaining a user-friendly interface that simplifies the customization process. This dynamic adaptability resolves the contradiction between ease of operation and problem-specific adaptability.
Data Source
AI summary
A machine learning application is selected from a plurality of machine learning applications. Each machine learning application corresponds to a different industry problem and includes standard features and machine learning pipelines specific to the corresponding industrial problem. The machine learning application receives a dataset for generating a model for making a prediction for the industrial problem corresponding to the selected machine learning application. The standard features are provided for display for the user to map variables in the dataset to the standard features. Mapping by the user is received through the user interface. The machine learning pipelines are applied to the dataset to train a plurality of models based at least on the mapping. The trained models are ranked and one of the trained models is selected based on the ranking. The selected trained model is to be used for making the prediction based on new data.