Enhancing Logistic Regression Using Neurochaos Learning

Abstract The rapid expansion of machine learning has intensified the demand for classification algorithms that are both effective and interpretable. Neurochaos Learning (NL), a chaos-based machine learning framework, has demonstrated strong classification performance; however, its computational complexity can limit practical applicability. To overcome this limitation, simplified NL variants have been introduced, employing chaos-inspired features such as the Tracemean and Fluctuation Index, which retain essential dynamical characteristics while reducing computational overhead. In this work, we integrate these simplified Neurochaos features with logistic regression, a classical and interpretable classification model. Logistic regression provides a mathematically transparent framework for evaluating the discriminative power and linear separability of chaos-derived features. The proposed approach is evaluated on several benchmark datasets, including Iris, Breast Cancer, Wine, Statlog, and Penguins datasets. Through systematic experimentation, we investigate the contribution of simplified NL features to classification accuracy, generalization capability, and stability across datasets. The results demonstrate that combining simplified Neurochaos features with logistic regression yields competitive performance while maintaining low computational complexity and high interpretability, making the approach suitable for practical and explainable machine learning applications. Keywords: Tracemean, Fluctuation index, Logistic regression, Neurochaos learning