Page 86 - ISMCON souvenir 2021

ISMSCON - 2021

Conclusion : It is possible to transform skewed nutrient intakes to normality using ISU method.

Keywords: Measurement error model, usual intake, ISU method, 24-hr recall

OS60: SMOOTHING TECHNIQUES: A SPECIAL ATTENTION TO

LOCALLY WEIGHTED SCATTERPLOT SMOOTHING

Sourav Raj K , R. Amala 2
1
1PG student, Department of Biostatistics, JIPMER Puducherry, 605006, souravraj109@gmail.com
2Assistant Professor, Department of Biostatistics, JIPMER Puducherry, 605006, amalar.statistics@gmail.com

Abstract
Smoothing refers to estimating a smooth trend, usually by means of weighted averages of observations.
The term smooth is used because such averages tend to reduce randomness by allowing positive and
negative random effects to partially offset each other. The random method, simple moving average,
random walk, simple exponential, and exponential moving average are some of the methods used for
data smoothing. But one of the most extensively used smoothing approach in medical research is locally
weighted scatterplot smoothing (LOESS).

LOESS is a procedure for fitting a regression surface to data through multivariate smoothing. The
dependent variable is smoothed as a function of the independent variables in a moving fashion analogous
to how a moving average is computed for a time series and the procedure is an adaptation of iterated
weighted least squares.
LOESS is now vividly used in the development of scoring system for different medical scenarios, for
e.g., SAPS, APS, LODS, SOFA, etc. This paper gives the insight into the LOESS technique and how
scrupulously it can be used to identify cut points for a continuous variables, and way of scoring these
cut points using LOESS technique. A step-by-step approach and application based on a simulated data
using R.
Keywords: Smoothing, LOESS, scoring system, application to medical data

OS61: A MODIFIED APPROACH ON WEIGHT FUNCTION FOR

SELECTION OF CONSISTENT FEATURES

Soutik Halder, Jitendra R. Gawde, Rajashree Dey, Sunil K. Yadav, Sanjay D. Talole &
Atanu Bhattacharjee
Section of Biostatistics, Centre for Cancer Epidemiology, Tata Memorial Centre, Kharghar, Navi Mumbai 410210,
India (Email Id: haldersoutik2015@gmail.com)

Abstract

Variable selection is the key challenge for high dimensional data in clinical studies. Removing the
irrelevant features is one of the necessary tasks for modelling. However, lack of reproducibility due
to random selection of different training and validation datasets is a common drawback in the existing
techniques. Also, it may happen that the sample data favour or oppose a few variables erroneously
because of selection bias. In this study, we propose a modified approach on weight function in the
resampling method for feature selection. The new weights are allocated for all the features in the dataset
and candidate features are chosen by placing a threshold value for the features weight. The coefficient

84 CONFERENCE SOUVENIR