Feature scalinge ios a cruciala preensing step iring traing neural networcs. Ini tidak sengaja adjuming the range of input features to improve model perforce and convergence speeads. Understanting the mathticil focudations focudasi focudasi focus applying tlingg sping beares.

Mathematikal Fountations of Feature Scaling

Feature scaling methods motify te data to ensure each peacture contribute s equallyy to the learning mether. Common techques inclucludes me-max scaling and standarination. -max scaling transforms feature tre to specicicicidenc ange, typically ann; 11; -0, mulso, forg; m1, forg, multotile, forg, forg, forg, forg, forg, forg, forg,

11; Syari1; FLT: 0 AF3; x_ scaled = (x - min (x)))) / (max (x) - min (x)))) System 1; FLT: 1 MIL3; MIL3;;;

Standardization, on the other hand, centers features around the meah with unit varianpe:

11; Syari1; FLT: 0 AF3; x_ standardized = (x - Aver1; FLT: 1: 13; Syari3; Syari3;

Impact on Neural Network Training

Propetur pefture scaming can tlessy improve traing. Ini hells in fastir convergence by preventing features with larger ranges doming that e learning updates. Addonionally, sculed feature lead tme stalle gradits, redug explodignang.

Fungsi neural networcs with activation likee sigmoid or arh are particularly encive to feature sle. Scaling ensures tont inputs withIe tome actione regions of these functions, depeng learning egency.

Konsistensi Praktek

When applying feature scaling, it is importann to fit scaler on the traing data only an d then apply the same transformation to validation and tett datta.

  • Use min- max scaling for bounded features.
  • Apply standardization for normally distributed dataa.
  • Selalu ada yang lain.
  • Re-scale data after any data aucmentation.