Hybrid ML Demand Prediction Model for New Product Sales
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing demand prediction models for new products fail to adequately address uncertainty and complexity, leading to inaccurate sales volume forecasts and increased financial risks for companies.
Innovation Solution
A machine learning-based hybrid model using K-means and ANN for demand pattern prediction, combined with QRNN for total demand forecasting, to calculate specific demand for each time slot, thereby improving sales volume prediction accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional demand prediction models are used, then the prediction process is simple, but the prediction accuracy is insufficient due to inability to address uncertainty and complexity of new products
Solution Approach 1:
The prediction model is divided into three distinct modules: a first model for demand pattern prediction using K-means and ANN, a second model for total demand prediction using QRNN, and a third model for calculating specific demand by integrating the outputs of the first two models. This segmentation allows each module to specialize in specific aspects of the prediction task, improving overall accuracy while maintaining manageable complexity through modular design.
Solution Approach 2:
The patent employs a hybrid composite model that integrates multiple different machine learning approaches (K-means clustering, Artificial Neural Networks, and Quantile Regression Neural Networks) into a unified prediction system. This composite structure leverages the strengths of each individual method to handle the uncertainty and complexity of new product demand prediction more effectively than any single model could alone.
2Reliability
If a single prediction model is used, then the model structure is simple, but it cannot adequately control the uncertainty of demand for new products
Solution Approach 1:
The ensemble prediction system is segmented into three specialized models: the first model handles demand pattern classification, the second model predicts total demand with uncertainty quantification, and the third model integrates these to produce final predictions. This segmentation enables the system to address different aspects of demand uncertainty separately, improving reliability while keeping each individual model relatively simple.
Solution Approach 2:
The patent merges multiple prediction approaches into an integrated ensemble model where the outputs of the first model (demand pattern) and second model (total demand) are combined in the third model to calculate specific demand. This merging of multiple independent prediction streams allows the system to control demand uncertainty more effectively than any single model could achieve alone.
3Measurement precision
If complex hybrid models are used, then prediction accuracy improves, but the difficulty of implementation and operation increases
Solution Approach 1:
The complex hybrid model is divided into three operationally independent modules that can be developed, trained, and maintained separately. The first model handles pattern recognition, the second handles total demand prediction, and the third handles integration. This segmentation reduces implementation difficulty by allowing teams to work on each module independently while still achieving high prediction accuracy through their combined output.
Data Source
AI summary
The present disclosure retates to a method and device for prediction with a machine learning-based hybrid model. The prediction method includes: creating a first model for predicting a demand pattern for a combination of product features through K-means and ANN based on historical data; creating a second model for predicting a total demand for a period to be predicted using QRNN; predicting the demand pattern in the first model and the total demand in the second model by using features of the new product as input variables in the first model and the second model; creating a third model for calculating a specific demand for each time slot by reflecting the total demand predicted through the second model in the demand pattern calculated through the first model; and predicting a result using the third model.