Fractal Forest: A Novel Approach to Synthetic Data Classification with Enhanced Feature Engineering and Performance Metrics

This study investigates the effectiveness of incorporating fractal dimension as a novel feature in multiclass classification using a synthetic dataset generated with make_classification. Employing a Random Forest classifier, the research evaluates the influence of fractal dimension on model performance across key metrics, including accuracy, precision, recall, and F1-score. With 150 samples and 10 features divided into three classes, results indicate that the classifier achieved an overall accuracy of 80%, demonstrating balanced performance across classes. Specifically, Class 1 showed a precision of 0.86 and an F1-score of 0.80, while Class 2 displayed slightly lower precision (0.77) and recall (0.71), reflecting some misclassification challenges. Class 3 achieved the highest recall at 0.93 and an F1-score of 0.85, suggesting the classifier's strong capability in identifying this class accurately. Including fractal dimension as a feature is found to enhance feature diversity, improving the Random Forest model's predictive robustness and accuracy. Future work will involve comparative analysis to quantify fractal dimension's specific contribution to classification, potentially extending this approach to real-world datasets. This study highlights the promise of innovative feature engineering techniques in improving machine learning model accuracy, especially in complex multiclass tasks.