Bayesian Optimization for Optimal Machine Learning Model Selection

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing methods for detecting unauthorized activities in enterprise systems face challenges in real-time detection, model generation, and resource optimization, particularly due to data discrepancies and the complexity of machine learning model training.

Innovation Solution

A computing platform is employed to collect and analyze login activity data, generate initial and production datasets, perform data quality checks, and apply Bayesian optimization to determine optimal machine learning models, reducing feature redundancy and improving model generation efficiency.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If machine learning processing complexity is increased to improve detection accuracy, then unauthorized activity detection accuracy is improved, but data processing time and resource consumption increase

Engineering Contradiction:

Improveunauthorized activity detection accuracyVSAvoiddata processing time

Core Design Contradiction:

Measurement precisionVSLoss of time

Solution Approach 1:

The system performs preliminary data quality checks, feature engineering, and dataset generation before model training. By preparing data in advance and creating initial test datasets with proper feature extraction, the system reduces the time required during actual model training and deployment, resolving the contradiction between accuracy and processing time.

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The machine learning process is divided into distinct segments: data collection, feature identification, dataset generation, data quality checking, model training, and deployment. This segmentation allows each step to be optimized independently, enabling comprehensive data processing for accuracy while managing overall processing time through efficient step-wise execution.

Inventive Principle:

Principle #1Segmentation

2Reliability

If comprehensive data processing procedures are implemented to improve model quality, then model generation reliability is improved, but system complexity increases

Engineering Contradiction:

Improvemodel generation reliabilityVSAvoidsystem complexity

Core Design Contradiction:

ReliabilityVSDevice complexity

Solution Approach 1:

The computing platform performs multiple functions within a unified system: data collection, feature engineering, dataset generation, quality checking, and model training. By consolidating these diverse functions into a single multi-functional platform, the system achieves comprehensive data processing for reliable model generation while avoiding the complexity of multiple separate systems.

Inventive Principle:

Principle #6Universality (Multi-functionality)

Solution Approach 2:

The system dynamically adjusts parameters such as feature selection criteria, dataset splitting ratios, and quality check thresholds based on the specific requirements of each modeling task. This parameter flexibility allows the system to maintain high reliability through comprehensive processing while adapting to different scenarios, thereby managing overall system complexity.

Inventive Principle:

Principle #35Parameter changes

3Difficulty of detecting and measuring

If more features are extracted from login activity data to improve detection capability, then unauthorized activity detection capability is improved, but data processing complexity and time increase

Engineering Contradiction:

Improvedetection capabilityVSAvoiddata processing complexity

Core Design Contradiction:

Difficulty of detecting and measuringVSDevice complexity

Solution Approach 1:

The system extracts only the most relevant features from login activity data through systematic feature engineering and selection processes. By identifying and extracting key features such as login time, location, device information, and user behavior patterns, the system achieves high detection capability while avoiding the complexity of processing all possible data elements.

Inventive Principle:

Principle #2Taking out (Extraction)

Solution Approach 2:

Different features are assigned different levels of importance and processing intensity based on their relevance to unauthorized activity detection. Critical features receive more sophisticated processing and analysis, while less important features are processed more simply. This local quality approach optimizes detection capability while managing processing complexity efficiently.

Inventive Principle:

Principle #3Local quality

Data Source

PatentUS11531734B2Determining optimal machine learning models

Publication Date: 2022.12.20 BANK OF AMERICA CORP

AI summary

Aspects of the disclosure relate to determining optimal machine learning models. A computing platform may collect, via a network, data indicative of login activity to an enterprise resource. One or more initial features indicative of the login activity may be identified. Based on the one or more initial features, an initial test dataset and a test model may be generated. Then, the test model may be deployed in a production environment. Subsequently, the computing platform may identify one or more production feature vectors. Then, the computing platform may generate, based on the one or more production feature vectors, a training dataset. Subsequently, the computing platform may perform, for the training dataset, a data quality check. Then, the computing platform may predict, by applying Bayesian optimization to the training dataset, an optimal machine learning model. Subsequently, the computing platform may apply the optimal machine learning model to detect unauthorized activity.