Return Area Prediction Using Regularized Regression and Clustering
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for predicting the return area of event visitors are limited in accuracy, especially for new events or venues, due to insufficient data for machine learning applications, leading to overtraining and decreased prediction accuracy.
Innovation Solution
A return area prediction device using a regularized regression method to acquire a regression result for predicting the number of visitors in each return area, and specifying the nearest-neighbor cluster closest to this result among previously clustered event group clusters to predict visitor numbers based on the center of gravity of this cluster.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a machine learning method is applied using limited historical event data, then the prediction model can be trained, but overtraining occurs and prediction accuracy deteriorates
Solution Approach 1:
The patent changes the approach from direct machine learning prediction to a two-stage process: first using regularized regression to obtain a baseline prediction, then finding the nearest neighbor cluster to adjust the prediction. This parameter change in the prediction methodology prevents overtraining by not relying solely on limited training data patterns.
Solution Approach 2:
The patent introduces an intermediary approach by using regularized regression results as a baseline and then mediating this with cluster-based adjustments. The nearest neighbor cluster acts as an intermediary that refines the regression result, preventing direct overfitting to limited training data while maintaining prediction accuracy.
2Ease of manufacture
If traditional regression methods are used for prediction, then the model is simple to implement, but prediction accuracy is insufficient for new events or venues
Solution Approach 1:
The patent merges traditional regularized regression with cluster analysis to create a hybrid prediction system. The regression provides a simple baseline model while the cluster analysis adds sophistication for handling new events and venues, combining the advantages of both approaches.
Solution Approach 2:
The patent performs preliminary clustering of historical event data into event group clusters before making predictions. This preliminary action organizes the data structure in advance, enabling the system to quickly find the nearest neighbor cluster and apply appropriate adjustments without complex real-time computations.
3Quantity of substance
If only the number of people alighting at the nearest station is used for prediction, then the data requirement is minimal, but the prediction accuracy of return area is limited
Solution Approach 1:
The patent adds another dimension to the prediction by incorporating spatial clustering information (event group clusters with centers of gravity) in addition to the temporal dimension of alighting numbers. This dimensional expansion allows the system to leverage spatial patterns in visitor behavior without requiring large quantities of additional data.
Data Source
AI summary
A return area prediction device (10) includes a prediction unit (12) that acquires a regression result of prediction of a number of visitors in each return area for a target event from event information of the target event using a regularized regression method, specifies, based on the acquired regression result, a nearest-neighbor cluster closest to the regression result among event group clusters previously clustered, and predicts the number of visitors in each return area for the target event based on at least a position of a center of gravity of the nearest-neighbor cluster.