Optimizing data collection for public health decisions: a data mining approach
BMC Public Health
Optimizing data collection for public health decisions: a data mining approach
Susan N Partington 0 1
Vasil Papakroni 2
Tim Menzies 2
0 Regional Research Institute, West Virginia University , 886 Chestnut Ridge Road, 5th Floor, P.O. Box 6825, Morgantown, WV 26506-6825 , USA
1 Division of Animal and Nutritional Sciences, West Virginia University , Morgantown, WV , USA
2 Lane Department of Computer Sciences and Electrical Engineering, West Virginia University , Morgantown, WV , USA
Background: Collecting data can be cumbersome and expensive. Lack of relevant, accurate and timely data for research to inform policy may negatively impact public health. The aim of this study was to test if the careful removal of items from two community nutrition surveys guided by a data mining technique called feature selection, can (a) identify a reduced dataset, while (b) not damaging the signal inside that data. Methods: The Nutrition Environment Measures Surveys for stores (NEMS-S) and restaurants (NEMS-R) were completed on 885 retail food outlets in two counties in West Virginia between May and November of 2011. A reduced dataset was identified for each outlet type using feature selection. Coefficients from linear regression modeling were used to weight items in the reduced datasets. Weighted item values were summed with the error term to compute reduced item survey scores. Scores produced by the full survey were compared to the reduced item scores using a Wilcoxon rank-sum test. Results: Feature selection identified 9 store and 16 restaurant survey items as significant predictors of the score produced from the full survey. The linear regression models built from the reduced feature sets had R2 values of 92% and 94% for restaurant and grocery store data, respectively. Conclusions: While there are many potentially important variables in any domain, the most useful set may only be a small subset. The use of feature selection in the initial phase of data collection to identify the most influential variables may be a useful tool to greatly reduce the amount of data needed thereby reducing cost.
Community survey methods; Data mining; Data collection; Ecological and environmental concepts; Nutrition
-
Introduction
Ideally, public health policy should be informed by research,
assessments and surveillance [1]. These activities rely on
the availability of current and accurate data collected at
the both the individual- and community-levels [2]. The
cost of conducting health research has recently become
an important consideration due decreases in available
funding. In the United States, federal funding for
biomedical research as a percent of total health care expenditures
decreased from 11% to 2% from 1980 to 2010 [3].
This paper explores one approach for reducing research
costs by reducing the number of survey items on two
community nutrition assessment instruments. In principle,
the approach described here is quite general and could
be applied to reducing the amount of data needed to
assess outcomes across a wide variety of health research
questions.
Background
Collection of primary data is one of the most expensive
and time consuming aspects of any research study [4].
To ensure data integrity, the collection process must be
consistently monitored. After collection, information from
paper forms requires double entry by hand or machine
scanning followed by manual confirmation of scanner
accuracy. Electronic collection of data either in person or
over the internet requires the purchase or development of
software to collect the data and if deployed over the
internet, web-based tools and the resources to host them [5].
In all cases, data cleaning and validation is required [6].
Resources needed increase in proportion to the amount
of data to be collected and managed. Further even a
rigorously monitored data gathering process is error
prone. Transcription errors, recording errors, data entry
errors, and errors resulting from equipment malfunction
all have the potential to distort findings and compromise
results [7]. Minimizing the amount of data needed to
produce an accurate assessment minimizes research costs as
well as the risk of errors.
Data mining
Data mining techniques employ algorithms or learners
that can build prediction models. Such algorithms include
linear regression, decision tree learners, Bayes classifiers,
random forests and support vector machines among
others [8]. Within these learners, there is often a feature
selection algorithm that identifies elements within a
dataset that are useful in the prediction model. There
are many feature selection algorithms including stepwise
regression, principle component analysis [9] and
information gain [8]. Feature selection studies have found that
ranking of singleton variables (as in stepwise regression)
does not work as well as exploring the rankings of
combinations of variables. That is, if every variable were ranked
only by their independent associa (...truncated)