AI / MLOps Project

MindScope

An AI-powered mental health analytics platform that detects psychological risk using clinically validated PHQ-9 and GAD-7 screening tools, combined with machine learning models to classify and score depression and anxiety severity — delivering real-time insights with explainability.

MLOps Scikit-learn Joblib EDA Python Streamlit Mental Health AI Risk Classification
PHQ-9
Depression Screening Scale
GAD-7
Anxiety Screening Scale
ML
Risk Classification Engine
Project Overview

MindScope was built to address the gap between clinical mental health screening and accessible data-driven insights. The system collects PHQ-9 and GAD-7 questionnaire responses, runs them through trained ML classification models, and returns a severity score with a risk tier — providing clinicians and researchers with a fast, reproducible diagnostic support tool. The entire pipeline is deployed via MLOps practices for reliability and reproducibility.

Key Features
🧠

Dual-Scale Screening

Supports both PHQ-9 (depression) and GAD-7 (anxiety) clinical questionnaires with automated scoring logic matching clinical standards.

📊

Risk Tier Classification

ML model classifies responses into Minimal, Mild, Moderate, and Severe tiers using Scikit-learn ensemble methods with cross-validated accuracy.

🔍

Exploratory Data Analysis

Deep EDA pipeline uncovering patterns in mental health datasets — score distributions, demographic correlations, and symptom clustering.

⚙️

MLOps Pipeline

Model serialized with Joblib and deployed with version control, enabling consistent predictions across environments with minimal latency.

📈

Interactive Dashboard

Streamlit-powered interface for real-time risk assessment, visual score breakdowns, and exportable session reports.

🛡️

Explainability Layer

Feature importance visualization to show which symptom responses contributed most to the risk classification — building clinician trust.

How It Works
1

Questionnaire Input

User answers PHQ-9 and GAD-7 items through the Streamlit interface. Each question maps to a clinical severity scale (0–3 per item).

2

Data Preprocessing

Responses are normalized, validated, and transformed into a feature vector matching the training-time schema for consistent inference.

3

ML Risk Classification

The pre-trained Scikit-learn model (serialized via Joblib) processes the feature vector and returns a severity class with confidence probability.

4

Score Visualization

Results are displayed as an interactive risk card with score breakdown, severity tier, and clinically-aligned recommendation text.

5

EDA Insights Panel

Aggregated analytics panel shows population-level distributions, score heatmaps, and feature correlation plots for research use.

Tech Stack
Python
Scikit-learn
Joblib
Pandas
NumPy
Streamlit
Matplotlib
Seaborn
EDA
MLOps
Model Serialization
Risk Classification
PHQ-9 / GAD-7
Render (Deployment)