With the big data collected from continuous glucose monitoring device, a machine learning algorithm pipeline was built to translate complex glucose recordings into intuitive class labels to automate decision making for medical doctors.

An interactive dashboard was built to ease visualisation of raw data and the extracted features (web tool, source code).

Manuscript based on this project was accepted under the title - Stratification of patients with diabetes using continuous glucose monitoring profiles and machine learning - in the journal Health Data Science.

COVID-19 case identification from a dynamic point of view based on only symptom measurements, to facilitate early diagnosis when clinical testing is unavailable. Key discoveries include: fever early in the illness and diarrhoea later in the illness increase risks of diagnosing COVID-19, while abdominal pain, headache, muscle ache, runny nose and a sore throat reduce the diagnostic risk (source code).

The research manuscript was published under title - Identifying COVID-19 cases in outpatient settings in the journal Epidemiology & Infection.

Birth anthropometry of Asian babies and its socioeconomic exposures was analysed, gestational age and gender-specific birth anthropometry charts were developed and compared to the widely used Fenton chart, based on 52 220 Chinese, Indian and Malay infants, born from 1991-1997 and from 2010-2017 in Singapore.

The manuscript was submitted under title - Birth anthropometry among three Asian ethnic groups in Singapore – new growth charts.

Using a recent idea from the interpretable machine learning literature, some regression-based diagnostic methods were developed which are useful for detecting when different parts of a summary statistic vector contain conflicting information about the model parameters. Conflicts of this kind complicate summary statistic choice, and detecting them can be insightful about model deficiencies and guide model improvement.

Publication can be found under title - Detecting conflicting summary statistics in likelihood-free inference in journal Statistics and Computing

Combining linear mixed models and predictive projections, a Bayesian clustering technique was created. Consider a subset of data sharing some features, their predictive replicates using chosen random effects is set as the same within group, with the remainder being integrated out using the conditional prior. The Bayesian clustering algorithm enables optimisation in heterogeneity and number of groups. This approach gives flexibility in clustering using the random effects of interest.

Manuscript has been submitted under title - Bayesian clustering using random effects models and predictive projections.