## Heatmap: Classification Accuracies
### Overview
The image is a heatmap visualizing classification accuracies across four methods (TTPD, LR, CCS, MM) for 12 categories (e.g., cities, neg_cities, sp_en_trans, etc.). Accuracy values are represented numerically with uncertainty (±) and color-coded via a gradient from purple (0.0) to yellow (1.0).
---
### Components/Axes
- **X-axis (Methods)**: TTPD, LR, CCS, MM (left to right).
- **Y-axis (Categories)**:
1. cities
2. neg_cities
3. sp_en_trans
4. neg_sp_en_trans
5. inventors
6. neg_inventors
7. animal_class
8. neg_animal_class
9. element_symb
10. neg_element_symb
11. facts
12. neg_facts
- **Legend**: Color gradient from purple (0.0) to yellow (1.0), labeled "Classification accuracies."
---
### Detailed Analysis
#### Method Performance by Category:
1. **cities**:
- TTPD: 97 ± 1 (light yellow)
- LR: 100 ± 0 (bright yellow)
- CCS: 98 ± 4 (yellow)
- MM: 92 ± 1 (orange-yellow)
2. **neg_cities**:
- TTPD: 100 ± 0 (bright yellow)
- LR: 100 ± 0 (bright yellow)
- CCS: 98 ± 8 (yellow)
- MM: 100 ± 0 (bright yellow)
3. **sp_en_trans**:
- TTPD: 99 ± 0 (bright yellow)
- LR: 99 ± 1 (yellow)
- CCS: 92 ± 14 (orange)
- MM: 93 ± 1 (orange-yellow)
4. **neg_sp_en_trans**:
- TTPD: 96 ± 1 (yellow)
- LR: 99 ± 2 (yellow)
- CCS: 89 ± 19 (orange)
- MM: 76 ± 5 (orange)
5. **inventors**:
- TTPD: 92 ± 1 (orange)
- LR: 90 ± 2 (orange)
- CCS: 81 ± 12 (orange)
- MM: 83 ± 1 (orange)
6. **neg_inventors**:
- TTPD: 92 ± 1 (orange)
- LR: 90 ± 3 (orange)
- CCS: 81 ± 14 (orange)
- MM: 92 ± 0 (orange-yellow)
7. **animal_class**:
- TTPD: 98 ± 0 (bright yellow)
- LR: 99 ± 1 (yellow)
- CCS: 85 ± 20 (orange)
- MM: 99 ± 0 (bright yellow)
8. **neg_animal_class**:
- TTPD: 99 ± 0 (bright yellow)
- LR: 97 ± 3 (yellow)
- CCS: 89 ± 18 (orange)
- MM: 99 ± 0 (bright yellow)
9. **element_symb**:
- TTPD: 96 ± 1 (yellow)
- LR: 97 ± 1 (yellow)
- CCS: 83 ± 20 (orange)
- MM: 89 ± 1 (orange-yellow)
10. **neg_element_symb**:
- TTPD: 92 ± 2 (orange)
- LR: 89 ± 10 (orange)
- CCS: 78 ± 21 (orange)
- MM: 74 ± 2 (orange)
11. **facts**:
- TTPD: 86 ± 1 (orange)
- LR: 87 ± 1 (orange)
- CCS: 82 ± 16 (orange)
- MM: 80 ± 1 (orange)
12. **neg_facts**:
- TTPD: 75 ± 0 (orange)
- LR: 81 ± 2 (orange)
- CCS: 71 ± 9 (orange)
- MM: 72 ± 1 (orange)
---
### Key Observations
1. **High Accuracy**:
- LR achieves 100% accuracy on **cities** and **neg_cities**.
- TTPD and MM show near-perfect performance on **neg_cities** (100 ± 0).
2. **Low Accuracy**:
- **neg_facts** is the weakest category, with TTPD at 75 ± 0 and CCS at 71 ± 9.
- **neg_element_symb** and **neg_sp_en_trans** show significant drops in accuracy for CCS (89 ± 19 and 78 ± 21, respectively).
3. **Uncertainty**:
- CCS has the highest variability (e.g., 92 ± 14 for **sp_en_trans**, 81 ± 14 for **neg_inventors**).
- TTPD and MM generally have lower uncertainty (±1–2) compared to CCS (±4–21).
---
### Interpretation
- **Method Strengths**:
- **LR** excels in **neg_cities** and **cities**, suggesting robustness in handling these categories.
- **TTPD** performs consistently well across most categories but struggles with **neg_facts**.
- **CCS** shows high variability, particularly in **neg_sp_en_trans** and **neg_element_symb**, indicating potential overfitting or sensitivity to noise.
- **Category Challenges**:
- **neg_**-prefixed categories (e.g., neg_cities, neg_facts) generally have lower accuracies, suggesting these are harder to classify.
- **neg_element_symb** and **neg_sp_en_trans** are outliers with notably poor performance for CCS and MM.
- **Color Consistency**:
- Yellow dominates high-accuracy cells (e.g., 99 ± 0), while orange reflects lower accuracies (e.g., 75 ± 0). The legend aligns perfectly with these values.
---
### Conclusion
The heatmap reveals that **LR** and **TTPD** are the most reliable methods overall, while **CCS** exhibits inconsistent performance, particularly in negative categories. The data underscores the importance of method selection based on the target category, with **neg_facts** and **neg_element_symb** being the most challenging for all approaches.
