Software for handling missing data in medical questionnaires  
Copyright (C) 2012 -2014

Dr Lalit Garg , Nanyang Technological University, Singapore 
Dr  Justin Dauwels, Nanyang Technological University, Singapore 
Dr Arul Earnest, Duke-NUS Graduate medical School, Singapore  
Dr Leong Khai Pang, Tan Tock Seng Hospital, Singapore


Standard approaches

Mean Substitution (MS)
 
Weighted K nearest neighbor (wKNN) estimation

Iterated local least square method for missing value imputation (ItrLLSimpute)

Standard SVD

Column normalized SVD 


Tensor decomposition based approaches

Standard CP

Column normalized CP 

All executables and an example combined in a single zip file



Software is a number of standalone executable applications that has been implemented with Matlab v2011b, but it does not require Matlab. If Matlab is not already installed in your computer install the Matlab Compiler Runtime (MCR) 7.16 (for Matlab R2011b) for Windows 32-bit from http://www.nd.edu/~goodlab/MCRInstaller-7.16-Win.exe. The MCR installation guide is also available at Mathworks.


Download
Click on 'the selected approach' from above to download the executable files of the software. Save it into a folder in your computer. Alternatively to download the software forr all approaches together download the 'AllApproachesTogether.zip'.                                                                                                 

Note: if Matlab is not already installed on your computer, install the Matlab Compiler Runtime (MCR) 7.16 (for Matlab R2011b) for Windows 32-bit from here or http://www.nd.edu/~goodlab/MCRInstaller-7.16-Win.exe. The MCR installation guide is also available at Mathworks.
Installation
Save downloaded files in a folder in your PC. If you have downloaded 'AllApproachesTogether.zip', unzip it in a folder.

Note:  To uninstall the software just delete the exe files downloaded. If zip file is downloaded delete it and also delete the folder where the contents of zip file are extracted. 
How to use

Input: save input data in an MS-Excel file with the name 'InputData.xls' . Place it in the same folder together with executables downloaded.

Note: The name of input file is case sensitive. The Excel file should contain only data in different sheets and all columns and row entries should be the actual data entries and no header or other columns should be there in the Excel file.

Execute: Just doubleclick the .exe file of the approach of your choice. Once execution successfully completes it would create a 'CompleteData<NameOfApproach>.xls' file, where <NameOfApproach> would be replaced by the name of the approach you selected to impute missing data.  Also the software would create an error log (saving the DOS command window contents). 

Output:  A missing entry should be denoted by a negative value (eg '-9'). Run the executable file downloaded to impute the data. The results (complete data) would be saved in a new Excel file 'CompleteData......xls' created by the software. 


Reference: Details of each method and appropriate references are provided in following paper:

Garg L, Dauwels J, Earnest A, Pang L (2014). Tensor based methods for handling missing data in quality-of-life questionnaires. IEEE Journal of Biomedical and Health Informatics. Volume:18, Issue:5, Pages: 1571 - 1580.

Dauwels J, Garg L, Earnest A, Pang LK (2012). Tensor Factorizations for Missing Data Imputation in Medical Questionnaires, The 37th International Conference on Acoustics, Speech, and Signal Processing (ICASSP), Kyoto, Japan, March 25 - 30, 2012.