MLInsights

Upload any CSV or Excel dataset, explore it, clean it, train 30+ machine learning models, compare results, and download everything — PDF report, cleaned dataset, session metadata, and trained .pkl files. Built with Next.js 14 + Tailwind CSS on the frontend and FastAPI on the backend.

Features

• Dataset Insights — shape, dtypes, describe, correlation heatmap, value counts per column
• Missing Values — per-column stats, smart cleaning options (fill mean/median/mode, drop rows/cols, normalize)
• 30+ ML Models — regression, classification, clustering; auto-recommended based on your dataset
• Training Progress — live modal with per-model elapsed time and estimated total time
• Results — metrics table, feature importance chart, confusion matrix (classification)
• Compare — side-by-side bar chart across all trained models
• Download — PDF report, CSV/XLSX dataset, meta.json session data, and a ZIP of all .pkl model files

Models Covered

Task	Models
Classification	Logistic Regression, Decision Tree, Random Forest, Gradient Boosting, HistGradient Boosting, XGBoost, LightGBM, CatBoost, SVM (RBF/Linear), KNN, Naive Bayes, LDA, MLP, Stacking, AdaBoost, Bagging, Extra Trees
Regression	Linear, Ridge, Lasso, Elastic Net, Bayesian Ridge, Huber, Decision Tree, Random Forest, Gradient Boosting, XGBoost, LightGBM, CatBoost, SVR, KNN, MLP, Stacking
Clustering	KMeans, DBSCAN, Agglomerative, Spectral, BIRCH, OPTICS

Tech Stack

Frontend

• Next.js 14 (App Router, TypeScript)
• Tailwind CSS — dark theme
• Recharts — charts and heatmaps

Backend

• FastAPI — modular REST API
• scikit-learn, XGBoost, LightGBM, CatBoost — ML engines
• pandas + pyarrow — data I/O and session storage
• ReportLab — PDF generation
• Uvicorn — ASGI server

Project Structure

MLInsights/
├── backend/
│   ├── main.py                 # FastAPI app entry point + CORS
│   ├── requirements.txt
│   ├── runtime.txt             # Python version pin for Render
│   ├── render.yaml             # Render deployment config
│   ├── routers/
│   │   ├── upload.py           # Dataset upload + session creation
│   │   ├── insights.py         # Shape, dtypes, describe, correlations
│   │   ├── cleaning.py         # Missing value analysis + cleaning
│   │   ├── models.py           # Model recommendations
│   │   ├── training.py         # Model training + .pkl export
│   │   └── report.py           # PDF, CSV/XLSX, meta.json, models ZIP
│   └── utils/
│       ├── session_store.py    # File-based session management
│       ├── data_utils.py       # Pandas helpers
│       ├── ml_models.py        # Model registry (30+ models)
│       └── report_gen.py       # PDF builder
└── frontend/
    ├── app/
    │   ├── api/download/       # Next.js proxy route (fixes cross-origin filename)
    │   └── page.tsx            # Main single-page app
    ├── components/
    │   ├── FileUpload.tsx
    │   ├── DataInsights.tsx
    │   ├── DataCleaning.tsx
    │   ├── ModelSelection.tsx  # Model cards + training modal
    │   ├── TrainingResults.tsx
    │   └── ReportDownload.tsx  # Download cards (PDF, dataset, JSON, ZIP)
    └── lib/
        ├── api.ts              # API fetch helpers
        └── utils.ts

Getting Started

Prerequisites

• Node.js 18+
• Python 3.11+

1. Backend

cd backend
python -m venv .venv

# Windows
.venv\Scripts\activate
# macOS / Linux
source .venv/bin/activate

pip install -r requirements.txt

# Copy env file and edit if needed
cp .env.example .env

uvicorn main:app --host 0.0.0.0 --port 8000 --reload

API runs at http://localhost:8000. Docs at http://localhost:8000/docs.

2. Frontend

cd frontend
npm install

# Copy env file
cp .env.example .env.local
# Edit .env.local — set NEXT_PUBLIC_API_URL=http://localhost:8000

npm run dev

App runs at http://localhost:3000.

API Overview

POST /api/upload/                        Upload CSV or XLSX → returns session_id
GET  /api/insights/{session_id}/summary  Shape, dtypes, describe, value counts
GET  /api/insights/{session_id}/corr     Correlation matrix
GET  /api/cleaning/{session_id}/missing  Missing value stats per column
POST /api/cleaning/{session_id}/clean    Apply cleaning strategy
GET  /api/models/{session_id}/recommend  Auto-recommended models for the dataset
POST /api/training/{session_id}/train    Train selected models → returns results
GET  /api/report/{session_id}/pdf        Download PDF report
GET  /api/report/{session_id}/dataset    Download cleaned CSV or XLSX
GET  /api/report/{session_id}/meta       Download meta.json session data
GET  /api/report/{session_id}/models_zip Download ZIP of all .pkl model files

Environment Variables

Frontend — create frontend/.env.local:

NEXT_PUBLIC_API_URL=http://localhost:8000

Backend — create backend/.env:

ALLOWED_ORIGINS=http://localhost:3000
UPLOAD_DIR=backend/uploads

License

MIT