F1 score vs Accuracy — Machine learning Made Easy #codingdidi #accuracy #f1
In this Hindi tutorial, Akansha Yadav explains why evaluating a model solely with accuracy can be misleading, especially on imbalanced datasets. She defines accuracy, presents its formula (TP + TN / total predictions), and illustrates the problem with a disease‑detection example where only 10 out of 1,000 cases are positive, causing a 99% accuracy model to appear perfect while missing all diseased patients. She then introduces precision, recall, and the F1 score, describing how the F1 score balances precision (the proportion of predicted positives that are truly positive) and recall (the proportion of actual positives that are correctly identified). The simple F1 formula, which depends on precision and recall, is explained. Finally, she advises when to prefer accuracy—balanced‑class problems where each class is equally important—and when to prefer the F1 score—situations where missing positive cases is risky, such as fraud detection, cyber‑attacks, or disease detection. The video serves as a concise guide for practitioners to select the appropriate evaluation metric based on class distribution and business risk.
