WEBINAR: Balance selection: new insights on penalised regression models with compositional predictors [24-10-2023 CET: UTC/GMT +2 h]

26/09/2023

Dear CoDa-member,

We are glad to announce the forthcoming CoDa-webinar that we hope will be of interest to you. 

Looking forward to meet you there!

Best wishes,

The CoDa Association Council

---------------------------------------------------------------------------------------------------

 

Balance selection: new insights on penalised regression models with compositional predictors  

Jordi Saperas Riera

 

  • Day: 24-10-2023
  • Time: 12:00h (CET: UTC/GMT +2 h)
  • Link: https://us06web.zoom.us/j/88591669413?pwd=REZDc0tmK2VhNlkrUGJwY1FVNm93Zz09 
  • Speaker: Jordi Saperas Riera, PhD candidate, University of Girona, Spain
  • Moderator: Gianna Monti
  • Title: Balance selection: new insights on penalised regression models with compositional predictors
  • Abstract: The most popular regularised regression methods are ridge, lasso, and elastic net methods. All of them share the aim of reducing the complexity of the regression linear model. To achieve this goal, they introduce a critical element in the form of a penalty function for the regression coefficients. Usually, the penalty function is based on a norm of the vector of coefficients. Importantly, this norm plays a fundamental role in determining the relevance of the coefficients within the model. In regression models involving compositional covariates, coherence of the norm with Aitchison geometry is of crucial importance. This importance has been highlighted in recent papers on the topic where authors proposed penalty functions based on the Manhattan distance (L1-norm) and the Euclidean distance (L2-norm) applied to the vector of coefficients expressed as clr-vectors. The proposed method is based on a newly-defined compositional norm called 𝐿1 pairwise logratio. That is, a norm calculated using the pairwise logratio vector. This innovative method exhibits the potential to distinguish between balances that exert influence on the response variable and those that remain inert. We illustrate its performance with a real dataset.
  • Keywords:  Simplex, Aitchison geometry, Lasso regression, compositional norm

 

---------------------------------------------------------------------------------------------------

Please note that the webinar will be recorded. Feel free to forward this message to your colleagues who may be interested in attending.

Check your local time at https://www.timeanddate.com/worldclock/converter.html?iso=20231024T100000&p1=31&p2=179&p3=152&p4=101