MinimalAI: Brain Hemorrhage Detection in Images Through Minimalist Machine Learning Approach

Pattern classification encompasses a wide range of algorithms, such as Multi-Layer Perceptron, Support Vector Machine, K-Nearest Neighbors, Naïve Bayes, Adaboost, and Random Forest, catering to diverse applications. However, a new trend, eXplainable-Artificial Intelligence, aims to enhance the user-friendliness and understandability of Machine Learning algorithms. This study introduces a novel pattern classification approach, incorporating the Minimalist Machine Learning paradigm and the dMeans feature selection algorithm. It is evaluated against Multi-Layer Perceptron, Naïve Bayes, K-Nearest Neighbors, Support Vector Machine, Adaboost, and Random Forest classifiers for CT brain image classification. The dataset includes grayscale images divided into two categories: CT without Hemorrhage and CT with Intra-Ventricular Hemorrhage. Most models achieved 50–75% accuracy with sensitivity and specificity ranging from 58% to 86%. Notably, the proposed methodology achieved remarkable accuracy of 86.50%, matching the top-performing classifier and surpassing state-of-the-art algorithms. This performance is attributed to its simplicity and practicality, aligning with the trend of generating easily interpretable algorithms. Overall, the study contributes to eXplainable-Artificial Intelligence advancement by presenting a high-accuracy pattern classification methodology that outperforms existing algorithms in specificity. The methodology's practicality makes it suitable for real-world applications, promoting transparency and interpretability in Machine Learning algorithms.