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?