# Technical Document Extraction: Sepsis Prediction Model (Training) Confusion Matrix
## 1. Document Overview
This image is a technical visualization of a **Confusion Matrix** for a machine learning model titled **"Sepsis Prediction Model (Training)"**. It evaluates the performance of a binary classifier by comparing predicted labels against ground truth (true labels).
## 2. Component Isolation
### Header
* **Title:** Sepsis Prediction Model (Training)
### Main Chart (Heatmap)
* **Y-Axis (True Label):** Categories are "sepsis" (top) and "no sepsis" (bottom).
* **X-Axis (Predicted Label):** Categories are "sepsis" (left) and "no sepsis" (right).
* **Cell Content:** Each of the four quadrants contains a classification acronym, an absolute integer count, and a percentage representing the row-wise normalized value.
### Legend (Color Bar)
* **Placement:** Located on the far right of the image.
* **Scale:** A continuous gradient of blue shades.
* **Markers:** 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8.
* **Function:** Represents the normalized ratio (0.2 to 0.8+) of the predictions within each class.
---
## 3. Data Table Reconstruction
The following table transcribes the data points contained within the heatmap cells:
| True Label \ Predicted Label | sepsis (Predicted) | no sepsis (Predicted) |
| :--- | :--- | :--- |
| **sepsis (True)** | **TP (True Positive)**<br>Count: 682<br>Percentage: 74.29% | **FN (False Negative)**<br>Count: 236<br>Percentage: 25.71% |
| **no sepsis (True)** | **FP (False Positive)**<br>Count: 151,326<br>Percentage: 19.34% | **TN (True Negative)**<br>Count: 631,214<br>Percentage: 80.66% |
---
## 4. Trend and Statistical Analysis
### Spatial Grounding & Color Verification
* **Top-Left (TP):** Dark blue color. Matches the ~0.74 mark on the legend. This indicates a high success rate in identifying actual sepsis cases.
* **Bottom-Right (TN):** Darkest blue color. Matches the >0.8 mark on the legend. This indicates the highest success rate in identifying non-sepsis cases.
* **Top-Right (FN):** Lightest blue/grey. Matches the ~0.25 mark on the legend.
* **Bottom-Left (FP):** Very light blue. Matches the ~0.19 mark on the legend.
### Performance Summary
1. **Sensitivity (Recall):** The model correctly identifies **74.29%** of actual sepsis cases (682 out of 918 total sepsis cases).
2. **Specificity:** The model correctly identifies **80.66%** of non-sepsis cases (631,214 out of 782,540 total non-sepsis cases).
3. **Class Imbalance:** There is a massive class imbalance in the training set. There are only **918** total "sepsis" instances compared to **782,540** "no sepsis" instances.
4. **Error Analysis:**
* The model missed 236 sepsis cases (False Negatives).
* The model incorrectly flagged 151,326 healthy cases as sepsis (False Positives). Due to the high volume of "no sepsis" data, the absolute number of False Positives is significantly higher than the True Positives, despite the percentage (19.34%) being relatively low.
## 5. Language Declaration
The text in this image is entirely in **English**. No other languages were detected.
