An end-to-end Machine Learning application that classifies consumer complaints using an ensemble of NLP models. This project features a real-time web dashboard that communicates with a Python/Flask backend to provide live predictions and signal extraction.
The Neural Audit Terminal is designed for high-efficiency auditing. It uses three different AI engines to provide a consensus-based classification.
- The initial state of the Neural Audit Terminal before a scan is initiated.
- When all active engines (Logistic Regression, SVM, and Naive Bayes) agree on the classification, the system flags a high-confidence consensus.
- In cases where the models disagree or return low probability scores, the terminal alerts the auditor to perform a manual review.
- Ensemble Scoring: Uses Logistic Regression, SVM, and Naive Bayes simultaneously.
- Real-time Signal Extraction: Identifies key "Impact Words" that triggered the AI's decision.
- Consensus Engine: Automatically flags results as "High Consensus" or "Low" based on model agreement.
- Latency Tracking: Measures backend processing time in milliseconds.
app.py: The Flask server and prediction API that processes real-time requests.index.html: The Neural Audit Terminal UI, styled with Bootstrap and powered by vanilla JavaScript.Plotly.js: Provides logic for rendering live probability distributions and historical trends.Consumer complaint classification.ipynb: contains data cleaning, EDA and model training logic.*.pkl: Pre-trained Scikit-Learn pipelines and Label Encoders.
-
Clone the repository:
git clone <repo-url> cd <repo-folder>
-
Install dependencies:
pip install -r requirements.txt
-
Run the application:
python app.py # Access the Terminal: Open http://127.0.0.1:5000 in your browser.