# Technical Document Extraction: Sepsis Prediction Model Confusion Matrix
## 1. Header Information
* **Title:** Sepsis Prediction Model (Test)
* **Language:** English
## 2. Component Isolation
### A. Main Chart: Confusion Matrix
The image displays a 2x2 confusion matrix (heatmap) evaluating the performance of a sepsis prediction model on a test dataset.
* **Y-Axis (True Label):**
* Top Row: `sepsis`
* Bottom Row: `no sepsis`
* **X-Axis (Predicted Label):**
* Left Column: `sepsis`
* Right Column: `no sepsis`
### B. Legend (Color Bar)
* **Location:** Right side of the chart.
* **Type:** Sequential blue color scale.
* **Range:** 0.35 (lightest blue/white) to 0.65 (darkest blue).
* **Function:** Represents the normalized proportion (percentage) of the row-wise classification.
---
## 3. Data Table Reconstruction
The following table represents the raw counts and row-normalized percentages extracted from the matrix cells.
| True \ Predicted | sepsis (Predicted) | no sepsis (Predicted) | Row Totals (Calculated) |
| :--- | :--- | :--- | :--- |
| **sepsis (True)** | **TP (True Positive)**<br>Count: 1198<br>Percentage: 66.93% | **FN (False Negative)**<br>Count: 592<br>Percentage: 33.07% | 1,790 |
| **no sepsis (True)** | **FP (False Positive)**<br>Count: 676391<br>Percentage: 33.51% | **TN (True Negative)**<br>Count: 1342175<br>Percentage: 66.49% | 2,018,566 |
---
## 4. Trend and Statistical Analysis
### Spatial Grounding & Color Verification
* **Dark Blue Cells (Diagonal):** The cells for **TP** (Top-Left) and **TN** (Bottom-Right) are the darkest blue. According to the color bar, these correspond to values > 0.65. This indicates the model correctly identifies both classes approximately 66-67% of the time relative to their true occurrence.
* **Light Blue/White Cells (Off-Diagonal):** The cells for **FN** (Top-Right) and **FP** (Bottom-Left) are light blue/white. These correspond to values between 0.33 and 0.35 on the color bar.
### Key Performance Metrics (Derived)
* **Sensitivity (Recall):** 66.93% (The ability to correctly identify patients with sepsis).
* **Specificity:** 66.49% (The ability to correctly identify patients without sepsis).
* **Class Imbalance:** There is a massive disparity in the sample sizes. The "no sepsis" group (2,018,566) is significantly larger than the "sepsis" group (1,790), representing a highly imbalanced dataset typical of medical screening.
* **False Positive Volume:** While the percentage of False Positives is relatively low (33.51%), the absolute number is very high (676,391) due to the large "no sepsis" population.
## 5. Summary of Findings
The model demonstrates balanced performance in terms of percentages, achieving roughly **66-67% accuracy for both positive and negative classes**. However, in a practical clinical setting, the high number of False Positives (676,391) compared to True Positives (1,198) suggests a very low Precision (Positive Predictive Value), which would likely lead to significant "alarm fatigue" for clinicians.
