Feature Engineering System for Healthcare Predictive Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current machine learning implementations require resource-intensive and time-consuming data preprocessing for individual models, leading to repetitive and inaccurate data preparation, especially in healthcare where patient data varies significantly, necessitating re-training models frequently and resulting in less accurate generalized predictions.
Innovation Solution
A model management system that determines similarity between pre-stored models and new patient data, reusing pre-calculated features to generate machine learning models efficiently, reducing the need for re-training and enhancing prediction accuracy by leveraging pre-stored models and features.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If individually trained models are used for each patient, then prediction accuracy is improved, but resource consumption and time for data preprocessing increase significantly
Solution Approach 1:
The patent merges common data preparation steps into a shared process across multiple models. A parent data preparation graph is created that can be reused by multiple child models, combining the benefits of individualized prediction with resource efficiency through shared preprocessing logic.
Solution Approach 2:
The system performs data preparation in advance and stores the prepared data for later reuse. By pre-processing patient data and storing it in a prepared state, the system eliminates the need to repeat expensive preprocessing operations when deploying individual models, thus reducing resource consumption while maintaining accuracy.
2Measurement precision
If separately pre-processed data is used for individual models, then model accuracy is improved, but data preparation time increases
Solution Approach 1:
Data is pre-processed and stored in advance in a reusable format. The system performs expensive data preparation operations beforehand and caches the results, so when individual models need to be deployed, they can quickly retrieve pre-prepared data without repeating the time-consuming preprocessing steps.
Solution Approach 2:
The parent data preparation graph is designed to be universal and reusable across multiple child models. A single data preparation process serves multiple modeling purposes, allowing the same prepared data to be used for different individual models without repeating the preparation work.
3Use of energy by moving object
If generalized models trained on aggregate data are used, then resource consumption is reduced, but prediction accuracy decreases
Solution Approach 1:
The system segments the modeling process into a shared parent component and individual child components. The parent data preparation graph handles common preprocessing tasks that can be shared, while child models maintain individualized processing for patient-specific predictions, combining resource efficiency with accuracy.
Solution Approach 2:
The patent combines aggregate-level data preparation (shared across all patients) with individual-level model application. Common preprocessing steps are merged into a parent graph that serves all patients, while individual model predictions are still generated for each patient, achieving both resource efficiency and personalized accuracy.
4Adaptability or versatility
If ad-hoc feature engineering changes are made, then model adaptability is improved, but development time and complexity increase
Solution Approach 1:
The parent data preparation graph is designed to be universal and can serve multiple child models with different purposes. By creating a reusable template for data preparation, the system enables adaptability across different modeling scenarios without requiring separate ad-hoc engineering for each case, thus reducing development time and complexity.
Solution Approach 2:
The system uses copying of the parent data preparation graph to create child models. Instead of creating ad-hoc engineering solutions for each model, the proven data preparation process is copied and reused, maintaining adaptability while significantly reducing the time and complexity of development.
Data Source
AI summary
Example implementations described herein are directed to systems and methods for feature preparation that receives patient feature data and determines similarity of pre-stored models with the patient feature data. In an example implementation, a database of the pre-stored models is analyzed to assess similarity indicating that feature preparation of the pre-stored models is compatible with the patient feature data. For similarity indicative of feature preparation to be utilized, the feature preparation is conducted for the patient feature data based on the pre-stored model determined to be similar. The feature preparation retrieves reusable features associate with the similar pre-stored model, where the reusable features comprise pre-calculated features of the model. A machine learning model is generated using results of the feature preparation and patient feature data; and a prediction is provided using the machine learning model.