I'm working with a dataset of 9 features. While 8 of them have values ranging from 0-255, one feature has significantly different values. I'm using this dataset with an LSTM autoencoder for anomaly detection and had a couple of questions about scaling:
Although RobustScaler is recommended for handling outliers, I found that StandardScaler outperforms it in my tests. Could you explain why this might be the case?
I experimented with using RobustScaler for the divergent feature and StandardScaler for the rest. This approach seemed promising. Would you recommend this hybrid scaling method, or should I stick to one scaler for all features?
I'd appreciate your insights and guidance on the best approach.
Thank you for your time and expertise.