icss-feature-selection 0.1.2

A scikit-learn compatible feature selector using the Interval Chi-Square Score (ICSS) algorithm.

Interval Chi-Square Score (ICSS) Feature Selection

A Python implementation of the Interval Chi-Square Score (ICSS) feature selection algorithm for interval-valued data. This library provides a scikit-learn compatible feature selector that ranks and selects features based on their statistical independence from class labels.

Overview

The Interval Chi-Square Score (ICSS) is a feature selection technique designed specifically for interval-valued data, where each feature is represented as an interval $[a_i, b_i]$ rather than a single continuous value. This approach is particularly useful in domains where data naturally comes as ranges or intervals, such as:

• Temperature ranges in climate studies
• Price ranges in e-commerce
• Uncertainty quantification in measurements
• Survey response ranges

The ICSS algorithm evaluates the degree of independence between each interval-valued feature and the target class label using a chi-square based approach adapted for interval data.

Features

• ✨ Scikit-learn compatible API (fit, transform, fit_transform)
• 📊 Works with interval-valued features (shape: (n_samples, n_features, 2))
• 🎯 Selects top-k features based on ICSS scores
• 📈 Integrates seamlessly with scikit-learn pipelines
• 🧬 Based on peer-reviewed research

Installation

Via PyPI (Recommended)

pip install icss-feature-selection

See the PyPI package page for more information.

From Source

git clone https://github.com/vinaykumarngitub/icss-feature-selection.git
cd icss-feature-selection
pip install -e .

Requirements

• Python >= 3.8
• NumPy
• scikit-learn

Quick Start

Basic Usage

import numpy as np
from icss_feature_selection import ICSSFeatureSelector

# Create dummy interval data: (n_samples, n_features, 2)
# Each feature is represented as [lower_bound, upper_bound]
n_samples = 100
X = np.random.rand(n_samples, 4, 2)
X = np.sort(X, axis=2)  # Ensure [min, max] order

# Create labels correlated with some features
y = np.zeros(n_samples)
y[(X[:, 0, 0] > 0.1) & (X[:, 0, 1] < 0.2)] = 1
y[(X[:, 2, 0] > 0.7) & (X[:, 2, 1] < 0.9)] = 1

# Select top 2 features
selector = ICSSFeatureSelector(k=2)
selector.fit(X, y)

# Check ICSS scores for each feature
print("ICSS Scores:", selector.scores_)

# Transform data to keep only selected features
X_selected = selector.transform(X)
print("Selected features shape:", X_selected.shape)  # (100, 2, 2)

Using with Scikit-learn Pipelines

from sklearn.pipeline import Pipeline
from sklearn.preprocessing import StandardScaler

# Create a pipeline
pipeline = Pipeline([
    ('icss_select', ICSSFeatureSelector(k=2)),
    # Add your own preprocessing/classification steps here
])

# Fit and use the pipeline
pipeline.fit(X, y)
X_transformed = pipeline.transform(X)

Input Format

The feature selector expects input in the following format:

• X: 3D NumPy array of shape (n_samples, n_features, 2)

  • n_samples: Number of samples in the dataset
  • n_features: Number of interval-valued features
  • 2: Each feature is represented as [lower_bound, upper_bound]

• y: 1D NumPy array of shape (n_samples,)

  • Target class labels (discrete values for classification)

Example Data Creation

import numpy as np

# Method 1: Random intervals
n_samples, n_features = 100, 5
X = np.random.rand(n_samples, n_features, 2)
X = np.sort(X, axis=2)  # Ensure min <= max

# Method 2: From existing data with uncertainty
lower_bounds = np.random.rand(n_samples, n_features)
upper_bounds = lower_bounds + np.random.rand(n_samples, n_features) * 0.5
X = np.stack([lower_bounds, upper_bounds], axis=2)

# Class labels
y = np.random.randint(0, 3, n_samples)  # 3 classes

API Reference

ICSSFeatureSelector

class ICSSFeatureSelector(BaseEstimator, SelectorMixin)

Parameters

• k (int, default=10): Number of top features to select based on ICSS scores.

Attributes

• scores_ (ndarray of shape (n_features,)): ICSS score for each feature after fitting.

Methods

• fit(X, y): Compute ICSS scores for all features.
• transform(X): Select features with highest ICSS scores.
• fit_transform(X, y): Fit and transform in one step.
• get_support(indices=False): Get a boolean mask or indices of selected features.

Complete Example

Here's a complete example demonstrating the feature selection workflow:

import numpy as np
from icss_feature_selection import ICSSFeatureSelector

# 1. Create dummy interval data
n_samples = 100
X = np.random.rand(n_samples, 4, 2)
X = np.sort(X, axis=2)  # Ensure [min, max] order

# 2. Create labels correlated with features 0 and 2
y = np.zeros(n_samples)
y[(X[:, 0, 0] > 0.1) & (X[:, 0, 1] < 0.2)] = 1
y[(X[:, 2, 0] > 0.7) & (X[:, 2, 1] < 0.9)] = 1

print(f"Original X shape: {X.shape}")  # (100, 4, 2)

# 3. Use the ICSSFeatureSelector to select top 2 features
selector = ICSSFeatureSelector(k=2)

# 4. Fit the selector
selector.fit(X, y)

# 5. Check the scores (higher is better)
print(f"ICSS Scores: {selector.scores_}")

# 6. Transform the data
X_selected = selector.transform(X)
print(f"Transformed X shape: {X_selected.shape}")  # (100, 2, 2)

# 7. Get selected feature indices
support_indices = selector.get_support(indices=True)
print(f"Selected feature indices: {support_indices}")

print("\n✓ Successfully performed ICSS feature selection!")

Algorithm Details

The ICSS algorithm works as follows:

1. Reference Intervals: For each feature, extract all unique interval values from the dataset.

2. Similarity Kernel: Define a similarity measure between intervals: $$\text{sim}(I^q, I^r) = \begin{cases} 1 & \text{if } I^q \text{ contains } I^r \ 0 & \text{otherwise} \end{cases}$$

3. Contingency Table: Build a contingency table using the similarity kernel to count occurrences of each reference interval in each class.

4. Chi-Square Calculation: Compute the chi-square statistic adapted for interval data: $$I\chi^2 = \sum_{i,j} \frac{(O_{ij} - E_{ij})^2}{E_{ij}}$$ where $O_{ij}$ are observed frequencies and $E_{ij}$ are expected frequencies.

5. Feature Ranking: Rank features by their ICSS scores and select the top-k.

For more details, please refer to the research paper.

Research Paper

This implementation is based on the following peer-reviewed research:

Interval Chi-Square Score (ICSS): Feature Selection of Interval Valued Data

• Authors: Guru, D.S.; Vinay Kumar, N.
• Conference: Intelligent Systems Design and Applications (ISDA 2018)
• Published: April 14, 2019
• Publisher: Springer, Cham
• Book: Advances in Intelligent Systems and Computing, Vol. 941
• Pages: 579–590
• Print ISBN: 978-3-030-16659-5
• Online ISBN: 978-3-030-16660-1
• DOI: https://doi.org/10.1007/978-3-030-16660-1_67

Links

• 📄 Springer: https://link.springer.com/chapter/10.1007/978-3-030-16660-1_67
• 🔬 ResearchGate: https://www.researchgate.net/publication/332410223_Interval_Chi-Square_Score_ICSS_Feature_Selection_of_Interval_Valued_Data

Citation

If you use this implementation in your research, please cite the original paper:

@incollection{guru2020interval,
  title={Interval Chi-Square Score (ICSS): Feature Selection of Interval Valued Data},
  author={Guru, D.S. and Vinay Kumar, N.},
  booktitle={Intelligent Systems Design and Applications},
  pages={579--590},
  year={2020},
  publisher={Springer, Cham},
  isbn={978-3-030-16660-1},
  doi={10.1007/978-3-030-16660-1_67}
}

Testing

Run the test example to verify the installation:

python tests/test.py

Expected output:

Original X shape: (100, 4, 2)
ICSS Scores: [...]
Transformed X shape: (100, 2, 2)

✓ Successfully created and tested ICSSFeatureSelector!

Use Cases

• Climate Data: Select important temperature/humidity ranges for weather prediction
• Medical Data: Identify key biometric ranges that correlate with disease diagnosis
• Financial Data: Select relevant price ranges for stock trading systems
• Sensor Data: Choose important measurement ranges from multi-sensor systems
• Survey Analysis: Identify key response ranges in survey data

Implementation Notes

• The algorithm uses a similarity kernel where an interval $I^q$ "contains" interval $I^r$ if $q_- \leq r_- \text{ and } q_+ \geq r_+$
• ICSS scores are computed using the chi-square formula adapted for interval data
• Features with higher ICSS scores have stronger association with the class label
• The implementation is compatible with scikit-learn's feature selection interface

License

This project is licensed under the Apache License Version 2.0 - see the LICENSE file for details.

Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

• Related Work

For other feature selection techniques and interval data methods, see:

• scikit-learn's feature selection module
• Symbolic Data Analysis literature on interval-valued data

---

Last Updated: November 2025