Clustered Feature Engineering for Faster, More Accurate ML Training
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing machine learning algorithms for predicting outcomes rely on human-generated input features that are limited in breadth and granularity, leading to inefficient and slow training processes due to the inclusion of unhelpful data.
Innovation Solution
Utilize unsupervised vector quantization algorithms like k-means clustering to automatically partition data records into groupings with correlation scores, filtering out unhelpful data and providing predictive input features to machine learning algorithms.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If human-generated input features are used for machine learning algorithms, then the system is easier to implement, but the training process becomes slower and less accurate due to limited breadth and granularity of features
Solution Approach 1:
The system performs self-service by automatically generating input features through unsupervised vector quantization algorithms without requiring manual human intervention. The algorithm autonomously partitions data records into groupings and subgroups, generates correlation scores, and creates predictive input features, thereby eliminating the bottleneck of human feature engineering and significantly improving training speed.
Solution Approach 2:
The patent replaces the mechanical human process of feature generation with an automated computational system using unsupervised vector quantization algorithms. This substitution transforms the manual, time-consuming feature engineering process into an efficient algorithmic process that automatically creates high-quality input features, resolving the contradiction between ease of implementation and training efficiency.
2Reliability
If all data records are used for training, then more data is available for learning, but unhelpful data slows down the training process and reduces accuracy
Solution Approach 1:
The system extracts only the most predictive data patterns by using unsupervised vector quantization to partition data records into meaningful groupings and subgroups. By generating correlation scores that indicate predictive value, the system extracts and prioritizes helpful data while effectively filtering out unhelpful data, thereby improving prediction accuracy without the time penalty of processing all raw data.
Solution Approach 2:
The patent applies local quality by creating distinct groupings and subgroups with different correlation scores, where each subgroup represents a specific pattern with varying predictive value. This allows the system to treat different data segments differently, focusing computational resources on high-value subgroups while minimizing processing of low-value data, thus improving accuracy efficiently.
3Adaptability or versatility
If human users generate input features manually, then the process is simpler to understand, but the features are limited in breadth and granularity based on human constraints
Solution Approach 1:
The system achieves high adaptability and versatility through self-service automation, where the unsupervised vector quantization algorithm automatically discovers and generates a comprehensive set of input features without human constraints. The algorithm can process data at any granularity level and create numerous groupings and subgroups, far exceeding what human users can manually produce, while maintaining full automation.
Data Source
AI summary
A method of generating an input for a machine learning algorithm may include collecting data records. Each data record may include a plurality of categories of data. The method may include using vector quantization to partition the plurality of data records into a plurality of groupings. Each of the groupings may be based on one or more of the plurality of categories of data. The method may include generating a correlation score for each of the plurality of groupings. The correlation score may be indicative of whether a particular group is indicative of a given outcome.