Measurement Error Behavior Modelling

• Missing value imputation using Singular Value Decompostion (SVD) onto RES panel data
• Consider only common stores from Profile panel, RES panel and RMS data
• Exclude stores with 0 purchase across 24 periods and stores with less than 3 periods of purchases

• Total of 90 behavior variables from Profile panel and RES panel data
• Clustering stores based on 90 behavior variables using CLARA, resulted in 27 clusters based on silhouette
• Build Classifier based on random forest for 2 purpose:
  • To identify most influential variables to behavior clusters
  • To predict the behavior class for incoming new stores

Error in file(file, "rt") : cannot open the connection