Machine Learning Incident Classification for Cloud IT

The patent replaces the manual mechanical review process with an automated machine learning system. Logistic regression models and decision trees automatically analyze incident data, extract features, and determine incident-problem relationships without human intervention, thereby maintaining accuracy while eliminating time consumption.

The patent introduces machine learning models as intermediaries between incident data and problem identification. These models process incident features and generate predictions about incident-problem relationships, serving as an automated mediator that resolves the contradiction between accurate identification and time efficiency.

The patent performs preliminary actions by pre-training machine learning models on historical incident data before actual incident analysis is needed. The models are prepared in advance with extracted features and learned relationships, enabling them to quickly analyze new incidents without manual review, thus improving both understanding and response speed.

The patent substitutes the manual analytical process with automated machine learning inference. The models automatically understand incident causes and scope by analyzing incident features against learned patterns, eliminating the need for manual review while maintaining or improving understanding accuracy and significantly increasing productivity.

The patent segments the machine learning system into multiple independent logistic regression models, each trained on different subsets of incident data or different feature sets. This segmentation allows the system to achieve high prediction accuracy through ensemble methods while keeping individual model complexity manageable and enabling parallel processing.

The patent uses parameter changes by training multiple logistic regression models with different parameter configurations (different feature subsets, different training data samples). This approach improves prediction reliability through diversity in model parameters while maintaining the simplicity of the underlying logistic regression algorithm for each individual model.

The patent extracts only the most relevant features from incident data using machine learning models. Instead of analyzing all incident attributes equally, the system identifies and focuses on key features that have the strongest correlation with problems, thereby maintaining measurement precision while reducing the time and computational resources required for analysis.

The patent applies partial action by analyzing a subset of the most important incident features rather than all features. The machine learning models identify and focus on critical features that provide sufficient correlation analysis accuracy, eliminating the need to process every incident attribute and thus reducing time and computational overhead.