sff 1.0.0

Mo ta ngan

Feature Selection with Hybrid TFFS

Description

The get_list_feature_tffs_hybrid function is used for feature selection based on high occurrence frequency during the Random Forest construction process, while also supporting integration with traditional feature selection methods.

Syntax

get_list_feature_tffs_hybrid(df, number_of_runs, n_estimators, percent, type=None, percent_hybrid=None)

Parameters

Parameter	Data Type	Description
df	DataFrame	DataFrame containing input data (dependent variable in the first column).
number_of_runs	int	Number of Random Forest runs to determine important feature frequencies.
n_estimators	int	Number of trees in the Random Forest.
percent	float	Percentage of features selected based on the highest occurrence frequency.
type	str, optional	(Optional) Traditional feature selection method for integration.
percent_hybrid	float, optional	(Optional) Percentage of features retained after hybrid selection.

Valid Values for type

If type is used, one of the following feature selection methods can be chosen:

type Value	Feature Selection Method
"MI"	Mutual Information
"PC"	Pearson Correlation
"FS"	Fisher Score
"BW"	Backward Selection
"FW"	Forward Selection
"RC"	Recursive Feature Elimination (RFE)
"LS"	Lasso Regression

Usage

Using Random Forest Only for Feature Selection

import pandas as pd
from sff.app import get_list_feature_tffs_hybrid

# Sample DataFrame
data = {
    "Class": [0, 1, 0, 1, 0, 1, 0, 1, 0, 1],
    "Feature1": [5, 8, 6, 7, 5, 8, 6, 7, 5, 8],
    "Feature2": [1, 2, 3, 4, 5, 6, 7, 8, 9, 10],
    "Feature3": [10, 20, 10, 20, 10, 20, 10, 20, 10, 20]
}
df = pd.DataFrame(data)

# Run function
selected_features = get_list_feature_tffs_hybrid(df, number_of_runs=10, n_estimators=100, percent=20)
print("Selected Features:", selected_features)

Integrating with Mutual Information (MI)

selected_features = get_list_feature_tffs_hybrid(df, number_of_runs=10, n_estimators=100, percent=75, type="MI", percent_hybrid=50)
print("Selected Features:", selected_features)

Integrating with Recursive Feature Elimination (RFE)

selected_features = get_list_feature_tffs_hybrid(df, number_of_runs=15, n_estimators=200, percent=75, type="RC", percent_hybrid=50)
print("Selected Features:", selected_features)

Integrating with Mutual Information (MI) - Example Code

import pandas as pd
from sff.app import get_list_feature_tffs_hybrid

# Sample DataFrame
data = {
    "Class": [0, 1, 0, 1, 0, 1, 0, 1, 0, 1],
    "Feature1": [5, 8, 6, 7, 5, 8, 6, 7, 5, 8],
    "Feature2": [1, 2, 3, 4, 5, 6, 7, 8, 9, 10],
    "Feature3": [10, 20, 10, 20, 10, 20, 10, 20, 10, 20]
}
df = pd.DataFrame(data)

# Run function
selected_features = get_list_feature_tffs_hybrid(df, 5, 20, 75, "MI", 50)
print("Selected Features:", selected_features)

Notes

• If type is not provided, the function will use only Random Forest for feature selection.
• If type is provided, the function will integrate Random Forest with the specified traditional method for optimal feature selection.
• percent_hybrid is applicable only when type is used.
• Note: percent (4th parameter) must be greater than percent_hybrid.
• Note: df must have the class column as the first column.