Diagnostic with incomplete nominal/discrete data

Abstract

Missing values may be present in data without undermining its use for diagnostic / classification purposes but compromise application of readily available software. Surrogate entries can remedy the situation, although the outcome is generally unknown. Discretization of continuous attributes renders all data nominal and is helpful in dealing with missing values; particularly, no special handling is required for different attribute types. A number of classifiers exist or can be reformulated for this representation. Some classifiers can be reinvented as data completion methods. In this work the Decision Tree, Nearest Neighbour, and Naive Bayesian methods are demonstrated to have the required aptness. An approach is implemented whereby the entered missing values are not necessarily a close match of the true data; however, they intend to cause the least hindrance for classification. The proposed techniques find their application particularly in medical diagnostics. Where clinical data represents a number of related conditions, taking Cartesian product of class values of the underlying sub-problems allows narrowing down of the selection of missing value substitutes. Real-world data examples, some publically available, are enlisted for testing. The proposed and benchmark methods are compared by classifying the data before and after missing value imputation, indicating a significant improvement.

Data-analytically derived flexible HbA1c thresholds for type 2 diabetes mellitus diagnostic

“…Data on 300 attributes from 2,700 attendances of 900 patients have been collected in recent years. The dataset has been used in several data mining applications for identification of DM [33,34] and related conditions. [35,36] The temporal data underwent compression to instantiate patients instead of attendances by calculating longitudinal means or modes on the amount of available data.…”

“…[33] On a different occasion [34] we followed an approach on a par with the General Location Model [37] to set MV in then existing a version of the DiScRi data, which is appropriate for mixed attribute type domains consisting of continuous as well as of categorical features.…”

“…We recycle these ideas here and propose a method which is a hybridisation of the Hot Deck approach we previously evaluated and advanced. [33] A difficulty is though presented by some unknown outcomes that cannot be easily predicted, which is expected of classification, because of MV in attributes the decision is supposed to be based on. The Hot Deck approach, which draws similar instances to enter MV by example, allows to overcome this difficulty as the class attribute or any other are treated the same.…”

“…[34] More attributes, even though less strong, can be deployed to enhance decision by ensuring greater similarity between compared instances, which is indeed how the Hot Deck method works. Although previously [33] we proposed how to make use of this approach for all-categorical data domains, it is most suited for all-continuous data domains. The method advanced in this work is data type insensitive.…”

“…The initial order of selectors can be guessed using the domain knowledge and refined after data completion using the Information Gain (IG) criterion -the higher, the prior -but it is possible to evaluate individual feature IG even from incomplete data. [33] The strength of the scheme for the automated data entry derives from the circumstance that the selectors, apart from being well-known, also possess high IG. Particularly, all the mentioned features but gender are on the joint top 20% list where a feature is included if it has a sufficiently high IG at least in one of the mentioned problems, that is, individual T2DM, CVD and HT problems and the combined one.…”