## Heatmap: Classification accuracies ### Overview The image is a heatmap visualizing classification accuracy metrics across multiple algorithms and categories. The heatmap uses a color gradient from purple (low accuracy) to yellow (high accuracy), with numerical values and uncertainty ranges (±) embedded in each cell. The data is organized in a matrix format with rows representing categories and columns representing algorithms. ### Components/Axes - **Title**: "Classification accuracies" (top center) - **X-axis (Columns)**: Algorithms labeled as: - TTPD - LR - CCS - MM - **Y-axis (Rows)**: Categories labeled as: - cities_de - neg_cities_de - sp_en_trans_de - neg_sp_en_trans_de - inventors_de - neg_inventors_de - animal_class_de - neg_animal_class_de - element_symb_de - neg_element_symb_de - facts_de - neg_facts_de - **Legend**: Right-aligned colorbar with values from 0.0 (purple) to 1.0 (yellow), labeled "Classification accuracies" - **Color Gradient**: Purple → Red → Orange → Yellow (higher accuracy) ### Detailed Analysis #### Algorithm Performance 1. **TTPD**: - cities_de: 77 ± 2 (orange) - neg_cities_de: 100 ± 0 (yellow) - sp_en_trans_de: 93 ± 1 (orange) - neg_sp_en_trans_de: 92 ± 3 (orange) - inventors_de: 94 ± 0 (orange) - neg_inventors_de: 97 ± 1 (yellow) - animal_class_de: 82 ± 0 (orange) - neg_animal_class_de: 92 ± 2 (orange) - element_symb_de: 88 ± 0 (orange) - neg_element_symb_de: 81 ± 1 (orange) - facts_de: 75 ± 2 (orange) - neg_facts_de: 59 ± 2 (red) 2. **LR**: - cities_de: 97 ± 4 (yellow) - neg_cities_de: 100 ± 0 (yellow) - sp_en_trans_de: 72 ± 10 (red) - neg_sp_en_trans_de: 96 ± 1 (yellow) - inventors_de: 97 ± 2 (yellow) - neg_inventors_de: 93 ± 5 (orange) - animal_class_de: 86 ± 3 (orange) - neg_animal_class_de: 92 ± 5 (orange) - element_symb_de: 82 ± 7 (orange) - neg_element_symb_de: 93 ± 4 (yellow) - facts_de: 80 ± 3 (orange) - neg_facts_de: 79 ± 5 (orange) 3. **CCS**: - cities_de: 75 ± 20 (orange) - neg_cities_de: 78 ± 23 (orange) - sp_en_trans_de: 74 ± 21 (orange) - neg_sp_en_trans_de: 72 ± 21 (orange) - inventors_de: 80 ± 23 (orange) - neg_inventors_de: 80 ± 22 (orange) - animal_class_de: 71 ± 16 (orange) - neg_animal_class_de: 79 ± 17 (orange) - element_symb_de: 67 ± 19 (red) - neg_element_symb_de: 69 ± 16 (red) - facts_de: 63 ± 10 (red) - neg_facts_de: 65 ± 11 (red) 4. **MM**: - cities_de: 69 ± 2 (red) - neg_cities_de: 100 ± 0 (yellow) - sp_en_trans_de: 93 ± 1 (yellow) - neg_sp_en_trans_de: 91 ± 4 (orange) - inventors_de: 96 ± 2 (yellow) - neg_inventors_de: 93 ± 3 (yellow) - animal_class_de: 81 ± 1 (orange) - neg_animal_class_de: 85 ± 2 (orange) - element_symb_de: 79 ± 4 (orange) - neg_element_symb_de: 70 ± 2 (red) - facts_de: 74 ± 0 (orange) - neg_facts_de: 59 ± 1 (red) ### Key Observations 1. **Highest Accuracy**: - LR and MM achieve 100% accuracy in neg_cities_de. - LR and MM also show 97-100% accuracy in inventors_de and neg_inventors_de. 2. **Lowest Accuracy**: - CCS struggles with neg_facts_de (65 ± 11) and element_symb_de (67 ± 19). - TTPD and MM have the lowest accuracy in neg_facts_de (59 ± 2 and 59 ± 1, respectively). 3. **Algorithm-Specific Trends**: - **LR**: Consistently high performance in neg_cities_de (100%) and inventors_de (97%). - **CCS**: Lower accuracy across most categories, with significant uncertainty (±10-23). - **MM**: Strong performance in neg_cities_de (100%) and inventors_de (96%), but weaker in neg_facts_de (59%). 4. **Category-Specific Trends**: - **neg_ categories**: Generally lower accuracy (e.g., neg_facts_de: 59-79%). - **de categories**: Higher accuracy in cities_de (77-97%) and inventors_de (94-97%). ### Interpretation The heatmap reveals that **LR and MM algorithms outperform TTPD and CCS** in most categories, particularly in neg_cities_de and inventors_de. The **neg_ categories** (e.g., neg_facts_de, neg_element_symb_de) consistently show lower accuracy, suggesting these may represent more challenging or underrepresented data. The **CCS algorithm** exhibits the highest variability (e.g., ±20 in cities_de), indicating potential instability in its performance. The **LR algorithm** demonstrates the most consistent high accuracy (93-100%) across categories, while **MM** balances strong performance in key areas with moderate results in others. The color gradient confirms that yellow cells (high accuracy) dominate for LR and MM, whereas red/orange cells (lower accuracy) are more prevalent for CCS and TTPD in neg_ categories.