🔍 AutoFeatureSelect

A lightweight, modular Python package for automatic feature selection in machine learning workflows.
It helps data scientists and ML engineers remove irrelevant, redundant, or noisy features to improve model performance, reduce overfitting, and simplify datasets — without manual guesswork.

---

✨ Why AutoFeatureSelect?

Most automated ML tools focus on feature generation, not feature elimination.

But throwing more features at a model often hurts performance. What if you want to cut the noise?

That's where AutoFeatureSelect shines:

• ✅ Selects only the most relevant features
• ✅ Works on any tabular dataset
• ✅ No need to generate new features
• ✅ Transparent, modular, and customizable
• ✅ Compatible with any ML framework (scikit-learn, XGBoost, LightGBM, etc.)

---

🚀 Installation

pip install smartfeatureselectml

---

📦 What This Package Does

AutoFeatureSelect provides a growing set of feature selection tools, including:

Method	Type	Description
low_variance_filter	Filter	Removes features with very low variance (near-constant columns)
correlation_filter	Filter	Drops features that are highly correlated with others
mutual_info_filter	Filter	Selects features with high mutual information with target
vif_filter	Filter	Removes multicollinear features using Variance Inflation Factor
rfe_filter	Wrapper	Uses model-based recursive elimination of weakest features
tree_importance_filter	Embedded	Uses tree model importance scores for feature selection
shap_filter	Embedded	Uses SHAP values to rank and select top features

---

📌 Functions and Their Parameters

🔹 filters.py

Function	Parameters
correlation_filter	df, target, threshold=0.9
low_variance_filter	df, target=None, threshold=0.01
mutual_info_filter	df, target, problem_type="classification", threshold=0.01
vif_filter	df, target=None, threshold=5.0, verbose=True

---

🔹 model_wrappers.py

Function	Parameters
tree_importance_filter	X, y, model=None, importance_threshold=0.01, top_n=None, verbose=True
rfe_filter	X, y, model=None, n_features_to_select=10, verbose=True
shap_filter	X, y, model=None, top_n=10, verbose=True

---

✅ Notes

• Only tree_importance_filter uses a threshold-like parameter, named importance_threshold.
• rfe_filter uses n_features_to_select to define how many features to retain.
• shap_filter uses top_n to keep the most impactful features based on SHAP values.

---

🔧 Example Usage

import pandas as pd
from sklearn.ensemble import RandomForestClassifier
from autofeatureselect.filters import (
    low_variance_filter,
    correlation_filter,
    mutual_info_filter,
    vif_filter
)
from autofeatureselect.model_wrappers import (
    tree_importance_filter,
    rfe_filter,
    shap_filter
)

# Load your dataset
df = pd.read_csv("data.csv")
target = "target_column"

# 1. Apply Low Variance Filter
df = low_variance_filter(df, threshold=0.01)

# 2. Apply Correlation Filter
df = correlation_filter(df, target=target, threshold=0.9)

# 3. Apply Mutual Information Filter
df = mutual_info_filter(df, target=target, problem_type="classification", threshold=0.01)

# 4. Apply VIF Filter
df = vif_filter(df, threshold=5.0)

# Separate features and target for model-based methods
X = df.drop(columns=[target])
y = df[target]

# 5. Apply Tree-Based Feature Importance
X = tree_importance_filter(X, y, model=RandomForestClassifier())

# 6. Apply RFE (Recursive Feature Elimination)
X = rfe_filter(X, y, model=RandomForestClassifier())

# 7. Apply SHAP-Based Selection
X = shap_filter(X, y, model=RandomForestClassifier())

# Combine X and y back
df_selected = pd.concat([X, y], axis=1)

# Final selected features
print(df_selected.head())

---

💡 When to Use Which Method?

Method	Best For	Notes
low_variance_filter	Removing constant or near-constant columns	Fast and simple
correlation_filter	Redundant features (highly correlated)	Pairwise comparison
mutual_info_filter	Non-linear relevance to target	Requires problem_type parameter
vif_filter	Multicollinearity	Good before linear models
rfe_filter	Model-driven selection	Slower but accurate
tree_importance_filter	Tree models (e.g. XGBoost, RandomForest)	Uses feature importances from trees
shap_filter	Global interpretability, model-agnostic	Uses SHAP values, slower but insightful

---

🔍 How to Explore Available Functions

To explore the available feature selection functions and understand their parameters, you can use built-in Python tools in any Python environment (terminal, IDE, or Jupyter Notebook).

---

📘 1. List All Functions in a Module

You can list all functions within a module using dir():

import autofeatureselect.filters as filters
import autofeatureselect.model_wrappers as models

# List functions in the filters module
print(dir(filters))

# List functions in the model_wrappers module
print(dir(models))

---

📑 2. Get Help on a Specific Function

Use the help() function to view the docstring of a specific function, including description, parameters, and return values:

from autofeatureselect.filters import mutual_info_filter

help(mutual_info_filter)

---

🔍 How It's Different from Other Tools

Feature	AutoFeatureSelect	Featuretools	tsfresh	feature-engine	scikit-learn
Focus	Feature Selection	Feature Generation	Feature Extraction	Manual Engineering	General ML
Auto-generate new features	❌	✅	✅	⚠️	⚠️
Remove irrelevant features	✅	❌	⚠️ Partial	✅	✅
Targeted data	Tabular	Relational	Time-series	Tabular	All
Easy to use	✅	❌	❌	⚠️	✅
Beginner-friendly	✅	❌	❌	⚠️	⚠️

---

🧪 Testing the Package

# Run from the root of the project directory
pytest test/

Ensure pytest is installed:

pip install pytest

---

📁 Project Structure

autofeatureselect/
│
├── filters.py # Statistical methods
│ ├── low_variance_filter()
│ ├── correlation_filter()
│ └── mutual_info_filter()
│
├── model_wrappers.py # Model-based methods
│ ├── tree_importance_filter()
│ ├── rfe_filter()
│ └── shap_filter()
│
└── init.py # Exposes selected functions

---

📘 Documentation

Every function is self-contained and comes with:

• Docstrings
• Clear parameters
• Defaults set to typical values
• Returns filtered DataFrame (with target column)

You can use Python’s built-in help:

from autofeatureselect.filters import mutual_info_filter
help(mutual_info_filter)

Or read the source — it’s clean and readable!

---

🤔 What It Does Not Do

• ❌ It does not create new features (use Featuretools, tsfresh, or scikit-learn for that)
• ❌ It does not automate the entire pipeline (no one-size-fits-all — feature selection must be task-specific!)
• ❌ It does not hide logic in “black boxes” — everything is transparent and user-controllable

---

🙌 Contributing

Want to add your own feature selection method?
PRs and discussions welcome! Just follow the modular style, and document everything clearly.

---

👤 Author

Shreenidhi TH
Developer passionate about building tools for applied ML and automation.

---

License

This project is licensed under the MIT License.

---