Interpretable machine learning framework for white blood cell classification using handcrafted morphological features with Chi square optimized selection

Blood cells, encompassing Red Blood Cells, White Blood Cells (WBCs), and platelets, are critical indicators of human health, reflecting physiological balance and immune responses. WBC classification is essential for diagnosing haematological disorders like leukaemia and infections, with distinct subtypes neutrophils, lymphocytes, monocytes, eosinophils, and basophils each fulfilling specialized immune roles. Traditional manual microscopy, though effective, is labour-intensive and prone to human error, often leading to delayed diagnoses. Conversely, automated methods, particularly those employing deep learning, require extensive datasets and suffer from limited interpretability, hindering their clinical adoption. This research introduces a robust machine learning pipeline that leverages handcrafted features to classify WBCs with high accuracy and transparency, addressing these challenges. By extracting a comprehensive set of statistical, colour, edge, shape, and texture features from blood cell images, followed by Chi-square based feature selection, the proposed approach reduces dimensionality while preserving discriminative information. The pipeline employs Random Forests, achieving an impressive accuracy of 97.2%, with precision and recall surpassing 96% across all WBC subtypes. These metrics demonstrate superior reliability compared to traditional microscopy and many deep learning models, particularly in resource constrained settings. The interpretability of handcrafted features allows clinicians to understand classification decisions, enhancing trust and usability in medical diagnostics. Additionally, the pipeline’s computational efficiency supports rapid processing, enabling real time clinical decision making and reducing diagnostic turnaround times. This method’s balanced performance, validated through stratified cross validation and statistical significance testing, ensures robustness across diverse datasets. By offering improved diagnostic accuracy, scalability, and clinical applicability, this pipeline positions itself as a valuable tool for haematological analysis, with potential extensions to other cell types and pathologies. Future work will focus on integrating deep learning for hybrid models and validating the system in real-world clinical environments.