[Paper Review] Fuzzy Artmap and Neural Network Approach to Online Processing of Inputs with Missing Values

Fulufhelo V. Nelwamondo, Tshilidzi Marwala|arXiv (Cornell University)|May 8, 2007

Fuzzy Logic and Control Systems13 references3 citations

TL;DR

This paper proposes an ensemble-based, online method for handling missing input values in neural networks without imputation or prediction. Using Fuzzy ARTMAP for classification and multi-layer perceptrons for regression, the approach achieves up to 9% better performance in regression tasks compared to auto-associative networks with genetic algorithms, while eliminating the need for data estimation and reducing computational overhead.

ABSTRACT

An ensemble based approach for dealing with missing data, without predicting or imputing the missing values is proposed. This technique is suitable for online operations of neural networks and as a result, is used for online condition monitoring. The proposed technique is tested in both classification and regression problems. An ensemble of Fuzzy-ARTMAPs is used for classification whereas an ensemble of multi-layer perceptrons is used for the regression problem. Results obtained using this ensemble-based technique are compared to those obtained using a combination of auto-associative neural networks and genetic algorithms and findings show that this method can perform up to 9% better in regression problems. Another advantage of the proposed technique is that it eliminates the need for finding the best estimate of the data, and hence, saves time.

Motivation & Objective

• To develop a real-time, online method for processing inputs with missing values in neural networks.
• To eliminate the need for imputing or predicting missing data, reducing computational cost and estimation errors.
• To improve performance in both classification and regression tasks under missing data conditions.
• To enable efficient condition monitoring in dynamic, real-world environments.
• To compare the proposed ensemble approach with existing methods such as auto-associative networks and genetic algorithms.

Proposed method

• An ensemble of Fuzzy ARTMAP networks is used for classification tasks, allowing online learning and incremental processing of incomplete inputs.
• An ensemble of multi-layer perceptrons (MLPs) is employed for regression, trained to handle missing values directly without imputation.
• The method avoids data imputation by treating missing values as part of the input structure, preserving input uncertainty.
• The ensemble approach aggregates outputs from multiple models to improve robustness and generalization in the presence of missing data.
• The system is designed for online operation, supporting real-time condition monitoring applications.
• Performance is evaluated by comparing the ensemble method against a hybrid approach using auto-associative neural networks and genetic algorithms.

Experimental results

Research questions

• RQ1 Can an ensemble-based neural network approach effectively process inputs with missing values in real-time without imputation?
• RQ2 How does the performance of the proposed method compare to imputation-based approaches in regression and classification tasks?
• RQ3 Does eliminating the need for data estimation improve computational efficiency and model accuracy?
• RQ4 Can Fuzzy ARTMAP and MLP ensembles maintain high accuracy when inputs contain missing values?
• RQ5 What is the relative improvement of the proposed method over auto-associative networks combined with genetic algorithms?

Key findings

• The proposed method achieves up to 9% better performance in regression problems compared to the auto-associative neural network and genetic algorithm approach.
• The ensemble method eliminates the need for estimating or imputing missing data, reducing processing time and avoiding estimation errors.
• The approach is effective for online condition monitoring due to its incremental learning and real-time processing capabilities.
• Fuzzy ARTMAP ensembles provide robust classification performance even with incomplete inputs.
• MLP ensembles maintain high regression accuracy without requiring data imputation.
• The method demonstrates strong generalization and stability across both classification and regression tasks with missing inputs.