Target Prediction Framework Using SHAP Clustering and Transfer Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing target prediction techniques struggle to provide an optimal demand forecasting technique due to the diversity of demand patterns represented through time series data, leading to underfitting issues in clusters with limited data.
Innovation Solution
A target prediction method and framework that clusters items with similar influence of input variables using SHAP values, generates a prediction model for each cluster, and applies transfer learning to fine-tune models for each cluster, utilizing deep learning models like MLP or FFNN.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If a single prediction model is trained for all items, then the model training process is simple and efficient, but the model cannot capture the diversity of demand patterns and leads to underfitting in clusters with limited data
Solution Approach 1:
The patent segments the items into multiple clusters based on SHAP values, which capture the importance and direction of input variables. Each cluster is then modeled separately with its own prediction model, allowing the diversity of demand patterns to be captured while maintaining training efficiency through transfer learning.
Solution Approach 2:
The patent changes the modeling approach from a single universal model to multiple cluster-specific models. By adjusting the model parameters and structure for each cluster based on its characteristics (derived from SHAP values), the system achieves both efficiency and accuracy.
2Reliability
If cluster-specific prediction models are generated for each cluster, then the prediction accuracy for diverse demand patterns is improved, but the model complexity and data requirements increase
Solution Approach 1:
The patent uses a universal deep learning model architecture that serves multiple purposes: it first trains as a pre-trained model on all items, then adapts to become cluster-specific models. This multi-functional approach maintains accuracy while managing complexity through a unified model structure.
Solution Approach 2:
The patent performs preliminary training of a pre-trained model on the entire dataset before creating cluster-specific models. This preliminary action establishes a solid foundation that reduces the complexity of subsequent cluster-specific training, as each cluster model builds upon the pre-trained weights rather than training from scratch.
3Reliability
If transfer learning is applied for each cluster, then the underfitting issue in clusters with limited data is solved, but the computational resources and training time increase
Solution Approach 1:
The patent performs preliminary training of a pre-trained model on the entire dataset before creating cluster-specific models. This preliminary action establishes a solid foundation that reduces the computational resources and training time needed for subsequent cluster-specific training, as each cluster model builds upon the pre-trained weights rather than training from scratch.
Solution Approach 2:
The patent applies transfer learning with a focus on local quality by fine-tuning the pre-trained model parameters specifically for each cluster's characteristics. This allows the model to adapt to local cluster patterns while retaining the general knowledge from the pre-trained model, optimizing the balance between accuracy and computational resources.
Data Source
AI summary
Proposed are a target prediction method using pre-training and transfer learning, and a target prediction framework for performing the same, the method including a data input step of inputting prediction datasets related to targets for prediction, a base model training step of training deep learning models by using the prediction datasets input in the data input step, a cluster classification step of classifying the prediction datasets into a plurality of clusters by using SHapley Additive explanations (SHAP) values, and a transfer learning step of inputting the plurality of clusters into the respective deep learning models and retraining respective weights through transfer learning after the cluster classification step, thereby providing an optimal target prediction technique so as to satisfy the diversity of target patterns represented through time series data.