Create README.md

Author: Vishal-sys-code

Machine Learning Templates/Feature Engineering Template/README.md

@@ -0,0 +1,89 @@
+# **FEATURE ENGINEERING**
+This is the process of transforming the raw data into a format that is easier to understand and use.
+It has different process. Some of them are:
+1. **Feature Transformation**
+ a. *Missing Value Imputation*
+ b. *Handling Categorical Features*
+ c. *Outlier Detection*
+ d. *Scaling*
+ e. *Column Transformer*
+ f. *Function Transformer*
+ g. *Binning and Binarization*
+2. **Feature Construction**
+3. **Feature Splitting**
+
+## Feature Transformation
+Feature Transformation is the process of modifying or changing the original features of a dataset in order to improve the performance of the ML models.
+Some of the techniques used in feature transformation are:
+* Missing Value Imputation
+* Handling Categorical Features
+* Outlier Detection
+* Scaling
+
+**Missing Value Imputation :**
+To choose the right method is always important.
+Missing Value Imputation can be done by:
+* Using Mean or Median if the data is not too skewed
+* Using Median if the data is skewed or outliers
+* Using Mode Imputation for categorical data
+* Use Forward Fill or Backward Fill for time series data
+* KNN Imputation if relationship between missing values and other features is complex.
+
+**Handling Categorical Features :**
+Categorical Features can be handled by:
+* Label Encoding
+* One Hot Encoding
+* Ordinal Encoding
+
+In deep, Categorical Features can be handled by:
+* **Nominal Encoding:** One Hot Encoding, One Hot Encoding (Multiple Categories) , Mean Encoding
+* **Ordinal Encoding:** Label Encoding, Target Guided Ordinal Encoding
+
+**Outlier Detection :**
+Outliers can be treated by:
+* Trimming
+* Capping
+* Treating Outliers as a missing value
+* Discretization
+
+Outliers can be detected by:
+* **Visual Method:** Box Plot and Scatter Plot
+* **Statistical Method:** Z-Score(Standard Score), IQR (Inter-quartile Range), Winsorization.
+
+**Scaling :**
+Scaling can be done by:
+* Standardization
+* Normalization
+* Min Max Scaling
+
+**Column Transformer :**
+There is a scikit library called ColumnTransformer which can be used to apply different transformations to different columns.
+It is used to create and apply seperate transformers for numarical and categorical data.
+
+```python
+transformer = ColumnTransformer(transformers = [
+ ('tnf-1', SimpleImputer, ['column_name']),
+ ('tnf-2', OrdinalEncoder, categories = [['CN1', 'CN2']], ['column_name'])
+ .....
+ .....
+ .....
+])
+```
+
+**Function Transformer :**
+It is used to normally distribute the data. If your data is skewed, you can use this transformer to make it normally distributed.
+
+Some of the methods to do it: 
+* Log transformation
+* Reciprocal transformation
+* Square root transformation
+* Box-Cox transformation
+* Yeo-Johnson transformation
+
+In sklearn, Function Transformer can be used for: log, reciprocal, square root. Power Transformer can be used for Box-Cox and Yeo-Johnson transformation.
+
+**Binning and Binarization :**
+Some of the datasets do contain irregular values in number. There are two techniques for handling this numerical to the categorical data:
+* Binning or Discretization
+* Binarization
+* K-means Binning