Testing Robustness of Sensor-based Models

Robustness is a measure how sensitive a model is tosmaller or larger deviations of the input from theassumptions the model relies on, or the distribu-tion observed in the train data. These deviationsare called outliers and may considerably degradethe performance of the predictive models. Classicalrobust statistics introduces a range of measures ofrobustness: breakdown value, sensitivity curve, in-fluence function, maxbias curve, etc.

Download as PDF

Student Target Groups:

  • Students in ICE and Computer Science.)

Thesis Type:

  • Master Project / Master Thesis

Goals and tasks:

  • Literature review on robust statistics and itsapplication in the context of outlier detectionand analysis of predictive models.
  • Robustness analysis of several models operat-ing on various data sets.
  • Extend robustness measures to improve ro-bustness of predictive models and to detectoutliers in existing data.
  • Integration of the solution into an existingcloud-based service.
  • Summary of the results in a written report,oral presentation.

Recommended Prior Knowledge:

  • A good knowledge of data analysis methodsand statistics, enjoy working with real data
  • Interest to learn robust statistics (no priorknowledge is required), creative thinking.
  • Programming skills in Python.

Start:

  • As soon as possible

Contact: