Page 171 - Contributed Paper Session (CPS) - Volume 3

CPS1983 Chong N. et al.
policymakers for their use. The labour market indicators that track the state of
the economy include, but is not limited to, employment, unemployment,
income, job vacancies, labour turnover, retrenchment, employment conditions
and hours worked.
With increasing demands for more granular and timely statistics, MRSD
has leveraged on technology in recent years to implement various data
analytics initiatives to meet those demands.
In the following sections, the data analytics initiatives are explored in
greater detail, and how they have helped MRSD to improve operational
efficiency and data quality.

2. Methodology
In this section, the methodology behind the data analytics initiatives is
explained in greater detail.

a) Predictive Modelling
With limited resources and increasing demands for large amounts of data
at short notice, it is important to optimise resources to collect as much survey
data as possible. Survey data is collected online or through phone or face-to-
face interviews. Predictive modelling is done by making use of demographic
information such as age, gender and household composition of the
respondents. A random forest model can be trained with past information to
predict optimal timings to contact the selected respondents.
A random forest model is built from growing multiple decision trees, with
each tree depending on the values of a random predictor sampled
independently (James, Witten, Hastie & Tibshirani, 2013). At each split, only
one of randomly selected predictors are considered from the full set of
predictors (James et al., 2013). This method prevents high correlation among
trees from the influence of strong predictors (James et al., 2013). The outcome
is taken from the average of all predictions to reduce variance
(James et al., 2013).
Assuming each household has a responsible adult, the simplified decision
tree shows the most appropriate timing of contact. For instance, if a household
comprises a couple of working age, it would be recommended to only contact
them outside normal working hours. However, for a family with young
children, establishing contact with them during working hours has a higher
probability of success as an adult is more likely to be at home to care for the
children (Diagram 1).

160 | I S I W S C 2 0 1 9