AutoML Dashboard

An end-to-end Automated Machine Learning web application — upload any tabular dataset and get a production-ready model with zero code.

Overview

AutoML Dashboard automates the entire machine learning workflow for tabular data. From raw CSV upload to a downloadable trained model, the pipeline handles everything — preprocessing, model selection, hyperparameter tuning, and evaluation — through an interactive web interface.

No ML knowledge required. No code. Just upload and go.

---

Features

Training Pipeline

• Auto task detection — automatically detects Classification vs Regression

• Data Quality Report — scans for missing values, outliers (IQR), class imbalance, and target leakage

• 5-Model Zoo benchmarked via 3-fold cross-validation:

  Task	Models
  Classification	Logistic Regression, Random Forest, XGBoost, SVC, KNN
  Regression	Linear Regression, Random Forest, XGBoost, SVR, KNN

• Automated Hyperparameter Tuning via RandomizedSearchCV on the best model

• Feature Importance bar chart (top 15 features)

• Full Hold-out Evaluation on 20% test split:

  Classification	Regression
  Accuracy, F1-Score, ROC-AUC	RMSE, MAE, MSE, R²
  Confusion Matrix (normalised)	Actual vs Predicted scatter
  Per-class Classification Report	Residuals Distribution
  ROC Curve (binary problems)

• On-demand Model Download — model serialized in memory, nothing written to disk until you click Download

What-If Single Predictor

• Dynamic form auto-generated from the training schema
• Instant prediction with confidence scores
• Class probability bar chart (classification)

Batch Prediction

• Upload a .pkl model + any new dataset
• Automatic schema validation — catches missing columns before inference
• Prediction Analytics: summary table with count, percentage, and confidence stats
• Download results as CSV

---

Project Structure

AutoML/
├── app.py                  # Streamlit UI — Training, What-If, Batch tabs
├── automl_core.py          # Core ML pipeline (preprocessing, training, evaluation)
├── requirements.txt        # Python dependencies with version pins
├── .streamlit/
│   └── config.toml         # Streamlit dark theme & server config
├── LICENSE
└── .gitignore

---

Getting Started

Prerequisites

• Python 3.9 or higher
• pip

Installation

# 1. Clone the repository
git clone https://github.com/Lipranj14/AutoML.git
cd AutoML

# 2. Create and activate a virtual environment
python -m venv .venv

# Windows
.venv\Scripts\activate

# macOS / Linux
source .venv/bin/activate

# 3. Install dependencies
pip install -r requirements.txt

# 4. Launch the app
streamlit run app.py

---

Usage Guide

Step 1 — Train a Model

1. Go to the Training Pipeline tab
2. Upload a .csv or .xlsx dataset
3. Select the target column
4. Click Start Auto-ML Pipeline
5. Review the leaderboard, metrics, and evaluation charts
6. Click Download Versioned Model (.pkl) to save the model

Step 2 — Run Batch Predictions

1. Go to the Batch Prediction tab
2. Upload the .pkl model you downloaded
3. Upload a new dataset with the same feature columns
4. Click Generate Predictions
5. Download the predictions as CSV

Step 3 — What-If Analysis

1. After training, go to the What-If Predictor tab
2. Enter individual feature values
3. Click Predict for an instant result with confidence

---

Deployment

This app is deployed on Streamlit Community Cloud

Tech Stack

Layer	Technology
Frontend / UI	Streamlit
ML Framework	scikit-learn, XGBoost
Data Processing	pandas, NumPy
Visualization	Plotly
Model Serialization	joblib

---

Supported Formats

Input	Accepted Formats
Training / Prediction Data	.csv, .xlsx
Model File	.pkl
Prediction Output	.csv (download)

---

---

Made with Streamlit & scikit-learn