Adaptive imputation of missing values for incomplete pattern classification

Abstract: In classification of incomplete pattern, the missing values can either play a crucial role in the class determination, or have only little influence (or eventually none) on the classification results according to the context. We propose a credal classification method for incomplete pattern with adaptive imputation of missing values based on belief function theory. At first, we try to classify the object (incomplete pattern) based only on the available attribute values. As underlying principle, we assume that the missing information is not crucial for the classification if a specific class for the object can be found using only the available information. In this case, the object is committed to this particular class. However, if the object cannot be classified without ambiguity, it means that the missing values play a main role for achieving an accurate classification. In this case, the missing values will be imputed based on the K-nearest neighbor (K-NN) and self-organizing map (SOM) techniques, and the edited pattern with the imputation is then classified. The (original or edited) pattern is respectively classified according to each training class, and the classification results represented by basic belief assignments are fused with proper combination rules for making the credal classification. The object is allowed to belong with different masses of belief to the specific classes and meta-classes (which are particular disjunctions of several single classes). The credal classification captures well the uncertainty and imprecision of classification, and reduces effectively the rate of misclassifications thanks to the introduction of meta-classes. The effectiveness of the proposed method with respect to other classical methods is demonstrated based on several experiments using artificial and real data sets.

Adaptive Imputation of Missing Values for Incomplete Pattern Classification

In the domain of pattern classification, the challenge of dealing with missing data is a persistent concern. The paper, Adaptive Imputation of Missing Values for Incomplete Pattern Classification, offers an innovative approach rooted in belief function theory to address the issue of incomplete data. This approach is noteworthy because it incorporates an adaptive method for imputing missing values, thus enhancing the precision of classification processes.

Overview of Methodology

The core of this research hinges on a credal classification framework integrated with belief function theory. This approach enables nuanced management of uncertainty and imprecision inherent in incomplete patterns. The method commences with tentative classification attempts using available attribute data. If successful classification is achieved, it is inferred that the missing values have negligible impact. Otherwise, missing values are estimated using a combined approach involving $K$-Nearest Neighbors (K-NN) and Self-Organizing Maps (SOM).

Adaptive Imputation Strategy

The adaptive imputation strategy delineated in this paper is vital for ensuring the reliability of classification results. For cases where full classification cannot be achieved without missing data, the SOM is implemented to optimize and reduce computational complexity by representing training classes with prototype vectors. The missing values are then estimated based on the closest matching prototypes identified within each class, leveraging the K-NN method.

Credal Classification Mechanism

Credal classification provides a comprehensive approach by allowing objects to belong to singleton classes and meta-classes with differing belief masses. By introducing meta-classes, this method effectively captures and reduces classification uncertainties from missing data misestimations. The use of belief function theory facilitates this approach through fusion rules, with the standard Dempster-Shafer (DS) rule being modified to preserve partial conflicting beliefs, thereby reflecting potential classification imprecision.

Empirical Results and Validation

The paper presents experimental validation using artificial and real data sets, comparing the proposed method against several classical methods such as KNNI, FCMI, and SOMI. Results demonstrate that the adaptive imputation approach produces lower misclassification rates, particularly in the context of high-dimensional and complex data sets. Computational efficiency is also highlighted, showing significant reductions compared to other methods.

Implications and Future Prospects

This research bears critical implications for practical applications in fields where data incompleteness is prevalent, such as sensor networks and medical diagnostics. The adaptive imputation method promises more robust classifiers under conditions of partial observability and incomplete data, offering enhanced decision-making capabilities in these sectors.

Looking forward, research in AI and pattern recognition could benefit from exploring further adaptations of belief function theory to incorporate dynamic changes in data availability and leveraging machine learning advancements for imputation strategies. The extension of credal classification to new domains, alongside improvements in algorithmic efficiency, suggests promising avenues for exploration and refinement.

In conclusion, adaptive imputation integrated with credal classification offers a substantial advancement in the treatment of incomplete pattern classification. By effectively managing the inherent uncertainties through belief function theory, this approach provides a viable path towards more accurate and reliable classification systems in applied sciences and engineering.