Transformation for continuous predictors



Hi all,
I aim to apply regression methods for now. So i have a highly skewed predictor( be it left skewed or right skewed). So regression method will be best suited for which case ?

1- transformed data using ladder of powers transformation(i.e log transformation, square transformation etc.)
2- converting the skewed data into normal distribution
3- transforming using ladder of powers and then converting into normal distribution.