Machine Learning Engineer for Small Island Developing States

Details

Mission and objectives

UNDP has a long-standing partnership with Small Island Developing States?(SIDS), with an estimated annual value of US$300m, supporting SIDS to advance national development priorities and respond to diverse challenges and opportunities.?UNDP’s?upgraded SIDS offer responds to their most pressing needs as well as greatest opportunities for accelerating sustainable development. Building on multipliers that promise to accelerate progress across the SDGs, and building on?UNDP’s?comparative advantage and specific expertise, UNDP is expanding its support in: Climate Action, developing Blue Economies and promoting Digital Transformation.

Context

- The next phase of the small island developing states (SIDS) data platform expansion involves the development of analytic and machine learning models to extract insight that might be deemed useful by a diverse group of end users. One of the first steps in this phase is the development of a data imputation machine learning sand-box interface. This interface was designed for the imputation of country-level development indicators (as categorized and visualized in the rest of the platform) where users can customize different imputation models, train them in real-time for a selected indicator and observe the resulting imputations/predictions along with other important outputs from the models. With a large number of indicators present and possibly missing a significant amount of values per year, assessing every prediction result or optimizing every model for every possible indicator is a highly challenging task requiring a collective effort.

Task description

- Run machine learning imputation model for specific indicators (for instance key indicators for SIDS) and assess the prediction/imputation results within the context of observed values. Assessments can range from visualizations to statistical tests to comparing performance to simple imputation methods - Suggest/test improvement in modelling approach at any stage (domain knowledge incorporation, preprocessing approach, hyperparameter tunning approach, modelling and selection, etc.) - Advance visualization of ML results (predictions and feature importance)