RATE meeting

November 5th

Previous meeting

Discussed some of the potential use cases at Achmea

Overview

  • IEEE VIS
  • Use cases
  • Implementation
Project Description Contact Position Keten Model Decision maker Benefi t of explanations ORV (automatic acceptance) Part of all customers are accepted automatically, others are forwarded to an expert. Senna van Iersel Cees Willemsen Data scientist Domain Analyst IT generiek Random Forest (R) Manual investigation team Reduce time spend investigating whether an ORV insurance request can be accepted or not. Health insurance churn prediction Peter (DS) helped Lizan (DA) to develop a model. DAs now re-train the same model every year. It is evaluated using AB testing. Peter Diks Lizan Onderwater-Kops Data scientist Domain Analyst ?? Random Forest Marketeers Help marketeers to send more effective emails/letters to prevent churn Pricing (customer damage insurance) 18 products, Multidisciplinary team. Risk analysis is the starting point. Delicate balancing act to determine the right price. Joost van Bruggen Senior manager Schade & Inkomen GLM DS/DA themselves - Understanding models - Allowing DS to use more complex models without trading explainability of the results. Recruitment (churn prediction internally) HR centraal. Predict whether employees are likely to churn, and use that for job application acceptance. Silke Lhoëst Ivo Vink Data scientist/Domain analyst Data scientist/Domain analyst HR Manual bivariate subgroup discovery (SAS) ?? More effective churn prevention by individual insights Wheel of knowledge 3 projects: - Value indication (solar panels) - Debtor model (inform callcenter) - Dynamic acceptance Only wants to spend minimal time, likes to show explanations to legal department fi rst. Multidisciplinary team. Would be really happy to see global rules . Geerte Cotteleer Data scientist (manager?) ?? DT / rules (?) ?? ?? Debtor management Junior DS, but in a multidisciplinary team, close to the DM. Prediction of Martijn Wagenaar Matthijs Dries Junior Data scientist Junior Data scientist ?? 2 Random Forests (R) 100s features 1M instances “less” features 100-150k instances Debtor team, sends reminders and debt collector - Per-customer debtor strategy - As a tool to explain to senior management ?? 20-30 features 1000-1500 instances Data: candidate/team/manager ~50 features (?) 705k instances (2017) Imbalance ~5 % Data: grouped website visits Data: Survey customers fi ll in Data

Use cases

Implementation

demo!

Deck November 5th

By iamdecode

Made with Slides.com

Deck November 5th

  • 28

More from iamdecode