About Innovations for Poverty Action: Innovations for Poverty Action (IPA) is a non-profit organization dedicated to discovering and promoting effective solutions to global poverty problems. In close partnership with decision makers, the policymakers, practitioners, investors, and donors working with the poor around the world, IPA designs and evaluates potential solutions to poverty problems using randomized evaluations, the most rigorous evaluation method available. We also mobilize and support these decision makers to use these solutions to build better programs and policies at scale. 

The following describes the work to be completed by a Research Consultant specialized in predictive modeling. The activities described will support the second phase of the IPA-funded pilot project “Using Predictive Modeling to Target Fraud Prevention in Uganda” (PIs: Matthew Bird and Rafe Mazer). 

These work activities are part of the “Development and Testing” phase of the pilot project for the objective: Develop predictive models of third-party fraud in digital finance in Uganda.

Activities:

• Use machine learning to develop predictive models of third-party fraud
  • Use MNO and UCC complaints data to identify fraud types and develop predictions using the k-fold cross-validation procedure; begin with LightGBM machine learning algorithm (variation of BoostedDecisionTree (GBDT), used in preliminary predictive modeling
  • Explore variations and extensions of predictive models based on different data sources, including, based on availability, MNO/FSP administrative and social media data to complement and contrast with official complaints data

The envisioned work will take place over a three-month period between late November 2021 and February 2022

Essential Duties and Responsibilities:

The Research Consultant (RA) will clean, analyse, and develop predictive models using existing admin (complaints) datasets and social media data. The consultant will also scrape additional relevant social media data for 2020-2021 and incorporate this into cleaning/analysis/modelling. Assignments include:

Phase 1: Data processing and descriptive analytics

1. Data processing and data cleaning, incorporate as available social media or other data
2. Exploratory Data Analysis (EDA) examining relationships between fraud complaints and channel, demographic variables, etc.
3. Identification of keywords (and most frequent) from text and all providers (general) and for each provider (specifically). (Do unigram, bigram, and trigram analysis)
4. Generate graphics of word distributions in transcriptions of all providers and for fraud complaints for each provider
5. Wordclouds for all providers collectively and each provider individually for fraud complaints
6. Top words most related to each other in the data for fraud complaints

Phase 2: Predictive Modeling

1. Pre-processing of data
2. Training, validation and comparison of clustering algorithms
3. Training, validation and comparison of predictive modeling algorithms

Deliverables include:

• Cleaned datasets of MNO complaints data and admin/social media data (including all data that were used in EDA and topic modelling in Phase 1). Data to be submitted in Excel format.
• Reports detailing Phase 1 analysis (EDA, keywords, graphics of word distributions, word clouds and topic modelling)
• Cleaned pre-processed datasets that were used for Phase 2 predictive modelling
• (if they differ from those submitted under Deliverable 1)
• Reports detailing Phase 2 analysis (predictive modelling)

Education and/or Work Experience Requirements:

• At least 3-4 years of experience working with quantitative data, including extensive experience cleaning, managing, analyzing large data sets, and running predictive models.
• Demonstrated experience running predictive models utilizing administrative and social media data
• Demonstrated experience scraping social media for use in data analysis
• High level of Python and/or R knowledge and Stata
• Master’s degree in computer science, statistics, economics, public policy, or related fields
• Excellent management and organizational skills
• Fluency in English with strong communication skills

Flexible, self-motivating, able to manage multiple tasks efficiently, and team player

Application process:

• Please complete the online assessment through this link: https://forms.gle/Lm8cDjzbSVSHqay48