🏥 Problem: Patient Status Analysis and Prediction

For this problem, you need to implement a machine learning model to predict the Status field using a provided dataset containing patient information. The dataset is organized in a CSV file with various features, and the model will be evaluated based on precision for the Dead class.

📊 Dataset

The dataset contains the following fields:

• Age: Patient's age.
• Race: Patient's race (e.g., White, Other).
• Marital Status: Patient's marital status.
• T Stage: Tumor stage (T1, T2, T3, T4).
• N Stage: Lymph node involvement.
• 6th Stage: Cancer stage according to the 6th edition of the TNM classification.
• differentiate: Tumor differentiation grade.
• Grade: Histological grade of the tumor (1, 2, 3, etc.).
• A Stage: Disease stage classification.
• Tumor Size: Tumor size.
• BMI: Body Mass Index.
• Heart Rate: Heart rate.
• Serum Creatinine: Serum creatinine level.
• Uric Acid: Uric acid level.
• Hemoglobin: Hemoglobin concentration.
• GFR (Glomerular Filtration Rate): Glomerular filtration rate.
• Serum Sodium: Serum sodium concentration.
• Serum Potassium: Serum potassium concentration.
• Serum Albumin: Serum albumin level.
• Lactate: Lactate concentration.
• Status: Patient status (Dead or Alive) – target field.

📝 Tasks

For the first subtasks, you will need to load the dataset and perform statistical analyses on test.csv.

Subtask 1 (10 points)

Classify kidney function for each patient in the test set based on GFR:

• Normal if GFR >= 90
• Mildly Decreased if 60 <= GFR < 90

Subtask 2 (10 points)

Compute the quartiles (Q1, Q2, Q3) of the Serum Creatinine column in the training set and classify patients in the test set as follows:

• Very Low if Serum Creatinine <= Q1
• Low if Q1 < Serum Creatinine <= Q2
• High if Q2 < Serum Creatinine <= Q3
• Very High if Serum Creatinine > Q3

Subtask 3 (10 points)

Determine BMI value:

1. Compute the median of the BMI column in the training set.
2. For patients in the test set, assign:

• 1 if BMI > median in train
• 0 if BMI <= median

Subtask 4 (10 points)

Count the number of patients in train with the same T Stage as the patient in test.

Subtask 5 (60 points)

Build an ML model to predict Status (Dead or Alive) based on features. Evaluation is based on precision for the Dead class.

🧮 Evaluation

The evaluation metric is precision for the Dead class:

where:

• TP = number of patients correctly predicted as Dead
• FP = number of patients incorrectly predicted as Dead

📌 Notes

• The target field is Status.
• Some fields may contain missing values (NaN) and need to be handled.
• Correlation analysis between features and the target variable is recommended.
• Detect and remove outliers if they affect model performance.
• Submitting a sample_output for local testing gives 5 points.

📄 Submission File Format

The submission.csv file must contain results for all 5 subtasks for each row in the test set.
Each test row generates 5 lines in the file, one for each subtask.

File structure:

subtaskID, datapointID, answer

Column meanings:

• subtaskID – subtask number (1–5)
• datapointID – patient ID from the test set (if no ID, use the row index)
• answer – subtask result:
  • Subtask 1: Normal / Mildly Decreased / Decreased
  • Subtask 2: Very Low / Low / High / Very High
  • Subtask 3: 0 or 1
  • Subtask 4: integer (count)
  • Subtask 5: 1 if model predicts Dead, 0 if model predicts Alive

Example for a single patient with ID 3220:

subtaskID datapointID answer
1 3220 Normal
2 3220 Low
3 3220 0
4 3220 5
5 3220 1

🗂️ Useful Resources

• Complete Starter Kit – contains a skeleton to start solving the problem