Tree-Based Probabilistic Forecasting With Leaf-Path Quantiles

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Conventional forecasting methods for ranges of values, such as ARIMA, neural networks, and Bayesian models, are computationally intensive, data-intensive, and require more run-time, failing to provide accurate probabilistic forecasts efficiently.

Innovation Solution

Utilizing tree-based models in a non-conventional manner to estimate the probability of forecasted values, including pre-processing historical data, tracking leaf paths, determining related historical samples, and computing quantiles, which allows for less data-intensive and faster probabilistic forecasting.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If conventional forecasting methods (ARIMA, neural networks, Bayesian models) are used to forecast a range of values, then probabilistic forecasts can be obtained, but computational resource consumption increases significantly

Engineering Contradiction:

Improveaccuracy of probabilistic forecastsVSAvoidcomputational resource consumption

Core Design Contradiction:

ReliabilityVSDevice complexity

Solution Approach 1:

The patent segments the ensemble of trained trees into multiple subsets, where each subset is used to forecast a specific quantile level. This segmentation allows the system to efficiently compute probabilistic forecasts by distributing the computational workload across different tree subsets, rather than requiring all trees to be processed for every quantile calculation.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The patent performs preliminary actions by pre-training multiple tree ensembles for different quantile levels during the offline training phase. This preliminary preparation stores the trained trees and their corresponding quantile mappings, enabling fast online inference without requiring complex real-time computations for probabilistic forecast generation.

Inventive Principle:

Principle #10Preliminary action

2Measurement precision

If conventional forecasting methods are used to forecast demand variation, then accurate probabilistic information is obtained, but data requirements increase

Engineering Contradiction:

Improveaccuracy of demand variation forecastingVSAvoiddata intensity

Core Design Contradiction:

Measurement precisionVSQuantity of substance

Solution Approach 1:

The patent employs dynamic tree ensembles that are trained on historical data to adaptively capture demand variation patterns. The system dynamically selects and combines trees from different ensembles based on the specific quantile level being forecasted, allowing efficient utilization of historical data without requiring excessively large datasets.

Inventive Principle:

Principle #15Dynamics

3Reliability

If conventional forecasting methods are used to produce probabilistic forecasts, then forecast ranges are obtained, but run-time increases

Engineering Contradiction:

Improveaccuracy of probabilistic forecastsVSAvoidrun-time for forecast generation

Core Design Contradiction:

ReliabilityVSLoss of time

Solution Approach 1:

The patent performs preliminary training of multiple tree ensembles offline, storing the trained trees and their quantile mappings. During online inference, the system quickly retrieves and applies the pre-trained trees to generate probabilistic forecasts, significantly reducing run-time compared to conventional methods that perform complex computations in real-time.

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The patent segments the forecast generation process by assigning different tree subsets to different quantile levels. This segmentation enables parallel processing and efficient retrieval during inference, reducing the time required to generate complete probabilistic forecasts across multiple quantile levels.

Inventive Principle:

Principle #1Segmentation

4Productivity

If tree-based models are used in conventional manner, then computational efficiency is achieved, but probabilistic forecasts cannot be provided

Engineering Contradiction:

Improvecomputational efficiencyVSAvoidcapability to provide probabilistic forecasts

Core Design Contradiction:

ProductivityVSReliability

Solution Approach 1:

The patent segments the ensemble of trained trees into multiple subsets, where each subset is dedicated to forecasting a specific quantile level. This segmentation enables tree-based models to provide probabilistic forecasts by selecting and applying the appropriate tree subset based on the desired quantile, maintaining computational efficiency while enabling probabilistic forecasting capability.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The patent creates multi-functional tree ensembles that can simultaneously provide both point forecasts and probabilistic forecasts across multiple quantile levels. By training trees on different quantile targets and organizing them into subsets, the system achieves universal applicability for various forecasting needs while maintaining the computational efficiency of tree-based models.

Inventive Principle:

Principle #6Universality (Multi-functionality)

Data Source

PatentUS12541739B2Systems and methods for probabilistic estimation in tree-based forecast models

Publication Date: 2026.02.03 KINAXIS INC

AI summary

Systems and methods in which a historical data set is pre-processed once per trained machine-learning model; a value of an unknown sample is forecast while tracking a leaf path of the unknown sample; the leaf path of the unknown sample is limited to a subset of trees in each trained-machine model; a set of related historical samples is determined based on the leaf path of the unknown sample, and a set of quantiles is determined from the leaf path of the unknown sample. Inventory is loaded according to the set of quantiles.