# Context This is a Glass Identification Data Set from UCI. It contains 10 attributes including id. The response is glass type(discrete 7 values) # Content Attribute Information: 1. Id number: 1 to 214 (removed from CSV file) 2. RI: refractive index 3. Na: Sodium (unit measurement: weight percent in corresponding oxide, as are attributes 4-10) 4. Mg: Magnesium 5. Al: Aluminum 6. Si: Silicon 7. K: Potassium 8. Ca: Calcium 9. Ba: Barium 10. Fe: Iron 11. Type of glass: (class attribute) -- 1 building_windows_float_processed -- 2 building_windows_non_float_processed -- 3 vehicle_windows_float_processed -- 4 vehicle_windows_non_float_processed (none in this database) -- 5 containers -- 6 tableware -- 7 headlamps # Acknowledgements https://archive.ics.uci.edu/ml/datasets/Glass+Identification Source: Creator: B. German Central Research Establishment Home Office Forensic Science Service Aldermaston, Reading, Berkshire RG7 4PN Donor: Vina Spiehler, Ph.D., DABFT Diagnostic Products Corporation (213) 776-0180 (ext 3014) # Inspiration Data exploration of this dataset reveals two important characteristics : 1) The variables are highly **corelated** with each other including the response variables: So which kind of ML algorithm is most suitable for this dataset Random Forest , KNN or other? Also since dataset is too small is there any chance of applying PCA or it should be completely avoided? 2) Highly **Skewed** Data: Is scaling sufficient or are there any other techniques which should be applied to normalize data? Like BOX-COX Power transformation?