📘 README for consumer_comp.py

Overview

composer_cpm.py is a machine learning pipeline for classifying consumer complaints into specific categories (e.g., credit reporting, debt collection, consumer loans, mortgages).

The script:

• Loads a consumer complaint dataset (CSV)
• Maps product categories to predefined labels
• Cleans and processes complaint text
• Extracts features using TF-IDF (with unigrams and bigrams)
• Trains and compares two models:
  • Logistic Regression
  • Multinomial Naive Bayes
• Evaluates models using accuracy, F1-score, and confusion matrix
• Saves the best model and an evaluation report
• Provides a prediction function for new complaints

---

🔧 Requirements

Install the dependencies before running the script:

pip install numpy pandas scikit-learn joblib matplotlib seaborn

---

📂 Input Data

• You must provide a CSV file of consumer complaints.
• Update the script variable CSV_PATH with the path to your dataset:

  CSV_PATH = "C:/Users/Asus/Desktop/complaints.csv"

Expected Columns

The script automatically detects a suitable text column. It prefers:

• Consumer complaint narrative
• consumer_complaint_narrative
• complaint_what_happened
• Complaint
  If not found, it falls back to Issue or Product.

The Product column is used to map complaints to one of 4 categories:

• 0 → Credit reporting, repair, or other
• 1 → Debt collection
• 2 → Consumer Loan
• 3 → Mortgage

---

▶️ How to Run

Run the script directly:

python consumer_comp.py

It will:

1. Load and clean the data
2. Train Logistic Regression and Naive Bayes models
3. Compare their performance
4. Save:
   • best_consumer_complaint_model.joblib → the best model
   • evaluation_report.json → evaluation metrics

---

📊 Outputs

1. Model performance (printed to console):

   • Accuracy
   • F1-macro score
   • Classification report
   • Confusion matrix

2. Saved files:

   • best_consumer_complaint_model.joblib (trained model)
   • evaluation_report.json (evaluation metrics)

---

🔮 Example Predictions

At the end of the run, the script tests a few sample complaints:

- Credit reporting, repair, or other -> I pulled my credit report and found an error ...
- Debt collection -> I received a call from a collector about a debt ...
- Consumer Loan -> I applied for a personal loan but the lender says ...
- Mortgage -> The mortgage servicer has been charging incorrect escrow ...

---

📌 Using the Saved Model

You can load the model and classify new texts using the provided function:

from composer_cpm import predict_texts

examples = [
    "My credit report shows incorrect information.",
    "A debt collector is harassing me."
]

predictions = predict_texts(examples)
print(predictions)
# ['Credit reporting, repair, or other', 'Debt collection']

---

🚀 Next Steps

• Expand LABEL_MAP to include more categories.
• Experiment with different models (e.g., SVM, Random Forest).
• Deploy the model as an API (Flask/FastAPI).