Skip to content

Latest commit

 

History

67 Commits

Folders and files

NameName
Last commit message
Last commit date
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

ANN Deep Learning Projects

Python TensorFlow Keras scikit-learn Streamlit GitHub Actions License: MIT

A structured portfolio of ten completed and deployed artificial neural network projects covering tabular classification, imbalanced learning, probability scoring, forecasting, optimization, computer vision, multi-output learning, and categorical embeddings.

Each project is developed as an end-to-end case study with reproducible source code, task-appropriate evaluation, reusable inference components, automated validation, responsible-use documentation, and a publicly accessible Streamlit application.

Portfolio status: 10 completed and deployed projects
Repository owner: Anmol Tripathi
Deployment portfolio: 10 Streamlit Community Cloud applications


Portfolio Objective

This repository demonstrates how artificial neural networks can be applied to practical business, analytical, risk, forecasting, optimization, and computer-vision problems.

Each project is designed to move beyond notebook-only experimentation and generally contains:

  • a clearly defined analytical or business problem;
  • reproducible data preparation and feature engineering;
  • an ANN architecture selected for the task;
  • training, validation, and test separation;
  • task-appropriate evaluation metrics;
  • saved preprocessing and model artifacts;
  • modular prediction and inference code;
  • automated tests or GitHub Actions validation;
  • an interactive public demonstration;
  • responsible-use guidance;
  • transparent limitations and future improvements.

The portfolio is intended to demonstrate skills relevant to Data Science, Machine Learning, Applied AI, Deep Learning, Data Analytics, Quality Analytics, Business Intelligence, Analytics Engineering, and AI application development.


Completed Projects

No. Project Neural-Network Problem Primary Deployment Status
1 Churn Classification Binary classification Streamlit Live Demo
2 Credit Card Fraud Detection Imbalanced binary classification Streamlit Live Demo
3 Credit Risk Probability Scoring Risk classification and probability scoring Streamlit Live Demo
4 Customer Lifetime Value Forecasting Multi-task regression and retention prediction Streamlit Live Demo
5 Diabetes Risk Screening Healthcare risk classification and probability scoring Streamlit Live Demo
6 Dynamic Pricing Optimization System Demand forecasting and constrained optimization Streamlit Live Demo
7 Handwritten Digit Recognition Multi-class computer vision Streamlit Live Demo
8 House Price Prediction Tabular regression and price estimation Streamlit Live Demo
9 Multi-Output Prediction System Mixed-output classification and regression Streamlit Live Demo
10 Tabular Deep Learning with Embeddings Classification with categorical embeddings Streamlit Live Demo

Portfolio at a Glance

Coverage Area Projects
Customer analytics Churn classification, customer lifetime value, multi-output prediction
Financial and risk analytics Fraud detection and credit-risk probability scoring
Healthcare analytics Diabetes risk screening
Regression and forecasting Customer lifetime value and house-price prediction
Decision optimization Dynamic pricing optimization
Computer vision Handwritten digit recognition
Multi-task and multi-output learning Customer lifetime value and multi-output prediction
Categorical representation learning Tabular deep learning with embeddings
Imbalanced learning Fraud detection and risk-oriented classification
Interactive deployment Ten Streamlit applications
Automated validation Project-specific tests and GitHub Actions workflows

What the Portfolio Covers

Customer and Business Analytics

  • Churn Classification identifies customers with an elevated likelihood of leaving.
  • Customer Lifetime Value Forecasting predicts future customer value and retention probability.
  • Multi-Output Prediction System produces churn, value, and engagement predictions through one shared neural network.
  • Dynamic Pricing Optimization System combines demand prediction with business-constrained price recommendations.

These projects demonstrate how predictive models can support retention, prioritization, customer segmentation, revenue planning, and operational decision-making.

Risk, Screening, and Anomaly Detection

  • Credit Card Fraud Detection focuses on severe class imbalance and minority-event detection.
  • Credit Risk Probability Scoring generates a predicted risk class and an interpretable probability score.
  • Diabetes Risk Screening demonstrates probability-based health-risk classification while clearly separating a portfolio prototype from clinical diagnosis.

These projects emphasize recall, precision, F1-score, ROC-AUC, PR-AUC, threshold selection, probability interpretation, and responsible communication of risk-oriented outputs.

Regression and Forecasting

  • House Price Prediction estimates continuous property values from structured attributes.
  • Customer Lifetime Value Forecasting predicts future monetary value.
  • Dynamic Pricing Optimization System forecasts demand and evaluates candidate pricing decisions.

These projects demonstrate numerical preprocessing, continuous-output ANN design, MAE, RMSE, R², residual analysis, and business interpretation of regression errors.

Computer Vision and Advanced Tabular Learning

  • Handwritten Digit Recognition applies convolutional neural-network concepts to multi-class image classification.
  • Tabular Deep Learning with Embeddings learns dense representations of categorical features and combines them with scaled numerical inputs.
  • Multi-Output Prediction System uses shared representation learning across classification and regression targets.

These projects expand the portfolio beyond standard sequential networks through image processing, the Keras Functional API, embeddings, shared layers, multiple inputs, and multiple outputs.


Project Summaries

01 — Churn Classification

Open Project 01 Live Demo

This project develops an ANN workflow for estimating customer churn risk from structured customer attributes. It connects preprocessing, probability-based prediction, evaluation, and stakeholder-facing interpretation in one deployable application.

Key capabilities: binary classification · structured preprocessing · probability scoring · threshold-aware evaluation · manual and batch inference · Streamlit deployment


02 — Credit Card Fraud Detection

Open Project 02 Live Demo

This project addresses fraud detection as a severely imbalanced binary-classification problem. Evaluation emphasizes minority-event detection and the trade-off between identifying suspicious transactions and controlling false alerts.

Key capabilities: imbalanced classification · minority-class evaluation · PR-AUC and ROC-AUC · probability interpretation · transaction scoring · interactive inference


03 — Credit Risk Probability Scoring

Open Project 03 Live Demo

This project builds a neural-network workflow for credit-risk classification and probability scoring. It translates model output into an interpretable risk signal while retaining clear limitations around consequential financial use.

Key capabilities: credit-risk modeling · class and probability outputs · threshold analysis · reusable inference · classification evaluation · responsible communication


04 — Customer Lifetime Value Forecasting

Open Project 04 Live Demo

This project applies multi-task neural learning to estimate future customer value and retention probability. Shared representations support related business outcomes while separate output heads preserve task-specific predictions.

Key capabilities: customer-value regression · retention classification · multi-task learning · categorical embeddings · output-specific evaluation · business interpretation


05 — Diabetes Risk Screening

Open Project 05 Live Demo

This project demonstrates ANN-based health-risk classification and probability scoring. It presents the output as an educational screening estimate—not a diagnosis—and emphasizes careful interpretation of healthcare predictions.

Key capabilities: health-risk classification · probability output · threshold evaluation · structured preprocessing · user-facing interpretation · non-diagnostic limitations


06 — Dynamic Pricing Optimization System

Open Project 06 Live Demo

This project combines neural demand prediction with constrained price evaluation. It demonstrates how predictive output can become a decision-support recommendation while respecting defined business boundaries.

Key capabilities: demand forecasting · candidate-price evaluation · constrained optimization · scenario comparison · recommendation explanation · what-if analysis


07 — Handwritten Digit Recognition

Open Project 07 Live Demo

This project applies neural-network and convolutional concepts to handwritten-digit image classification, from image preprocessing to class-probability inspection and interactive inference.

Key capabilities: multi-class image classification · image normalization · class probabilities · confusion-matrix evaluation · uploaded-image inference · computer-vision interface


08 — House Price Prediction

Open Project 08 Live Demo

This project develops an ANN regression workflow for estimating property values from structured housing attributes. It emphasizes regression-specific evaluation and interpretation.

Key capabilities: tabular regression · mixed-feature preprocessing · MAE, RMSE, and R² · residual analysis · property-level inference · interactive price estimation


09 — Multi-Output Prediction System

Open Project 09 Live Demo

This project uses a shared neural representation to generate multiple customer predictions, including classification and regression outputs, with target-specific training and evaluation.

Key capabilities: shared representations · mixed-output prediction · multiple heads · target-specific losses and metrics · consolidated scoring · reusable inference


10 — Tabular Deep Learning with Embeddings

Open Project 10 Live Demo

This project learns dense representations for categorical variables and combines them with scaled numerical features through the Keras Functional API.

Key capabilities: categorical embeddings · multiple inputs · numerical integration · Keras Functional API · probability scoring · reusable tabular inference


ANN Architecture Coverage

Architecture or Pattern Demonstrated Through
Feed-forward neural network Churn, fraud, credit risk, diabetes, and house-price projects
Binary classification head Churn, fraud, credit risk, diabetes, and tabular embeddings
Multi-class classification head Handwritten digit recognition
Regression head House-price prediction and customer lifetime value
Multi-task architecture Customer lifetime value forecasting
Multi-output architecture Multi-output prediction system
Shared hidden representation Customer lifetime value and multi-output prediction
Categorical embeddings Customer lifetime value and tabular deep learning
Convolutional image processing Handwritten digit recognition
Predictive model with optimization layer Dynamic pricing optimization

Evaluation Coverage

Task Evaluation Methods
Standard binary classification Accuracy, precision, recall, F1, ROC-AUC, confusion matrix
Imbalanced classification Minority-class recall, precision, F1, PR-AUC, threshold analysis
Probability scoring Probability distributions, discrimination metrics, threshold interpretation
Multi-class classification Accuracy, macro/weighted F1, class report, confusion matrix
Regression MAE, RMSE, R², residual analysis
Multi-task and multi-output learning Output-specific losses and metrics
Forecasting and optimization Prediction error, scenario comparison, constraint-aware recommendations

Why multiple evaluation methods matter

  • Accuracy alone can hide minority-class failures.
  • A probability score should not automatically be interpreted as certainty.
  • Regression quality requires magnitude-of-error and residual analysis.
  • Multi-output models must be evaluated separately for each target.
  • Optimization recommendations depend on both model quality and business constraints.
  • Healthcare and financial predictions require careful limitations and human oversight.

What the Repository Demonstrates

End-to-End Deep-Learning Delivery

The repository demonstrates problem definition, data validation, preprocessing, feature engineering, deterministic splitting, neural-network configuration, training, evaluation, error analysis, artifact preservation, reusable inference, testing, CI validation, application development, deployment, documentation, and responsible-use communication.

Reliable and Reusable Engineering

  • preprocessing fitted on training data only;
  • consistent feature order between training and prediction;
  • saved scalers, imputers, encoders, and metadata;
  • safe handling of missing values and unseen categories;
  • modular source files rather than notebook-only logic;
  • automated tests for important prediction paths;
  • project-specific GitHub Actions workflows;
  • deployment assets separated from training workflows;
  • large-file and secret protection through .gitignore.

Business Translation

Depending on the project, the applications provide risk levels, prediction probabilities, customer segments, value estimates, pricing recommendations, scenario comparisons, business interpretations, batch summaries, and downloadable scored datasets.

Responsible Model Communication

Each project documents its intended scope and limitations. Synthetic or sample data is identified where applicable, and the applications avoid presenting portfolio models as production-ready financial, medical, or operational decision systems.


Repository Convention

ann-deep-learning-projects/
├── .github/
│   └── workflows/
│       └── project-specific-ci.yml
├── 01-churn-classification/
├── 02-credit-card-fraud-detection/
├── 03-credit-risk-probability-scoring/
├── 04-customer-lifetime-value-forecasting/
├── 05-diabetes-prediction/
├── 06-dynamic-pricing-optimization/
├── 07-handwritten-digit-recognition/
├── 08-house-price-prediction/
├── 09-multi-output-prediction-system/
├── 10-tabular-deep-learning-with-embeddings/
├── .gitignore
├── LICENSE
├── PROJECT_ROADMAP.md
└── README.md

A typical project may contain app/, data/, images/, models/, notebooks/, outputs/, src/, tests/, a project README, dependencies, and supporting reports. Exact files differ by task, but reproducibility, modular code, evaluation, deployment, validation, documentation, and transparent limitations remain consistent.


Continuous Integration

The repository uses project-specific GitHub Actions workflows. Depending on the project, automated checks validate Python syntax, imports, tests, preprocessing and inference paths, model metadata, Streamlit entry points, documentation references, oversized files, and accidental secret inclusion.

Open GitHub Actions


Deployment Directory

All ten projects are presented through Streamlit Community Cloud. Deployment files are maintained inside the relevant project folders, normally under app/ or through project-level Streamlit entry points and requirements files. The live links are available in the Completed Projects table and each Project Summary.


Run a Project Locally

1. Clone the repository

git clone https://github.com/unit-mole/ann-deep-learning-projects.git
cd ann-deep-learning-projects

2. Enter a project

cd 01-churn-classification

Replace the folder name with the project you want to run.

3. Create a virtual environment

Windows

py -m venv .venv
.venv\Scripts\activate

macOS / Linux

python3 -m venv .venv
source .venv/bin/activate

4. Install project dependencies

python -m pip install --upgrade pip
python -m pip install -r requirements.txt

5. Follow the project README

Projects may run through Jupyter notebooks, Python scripts, Streamlit applications, manual inference forms, or batch CSV scoring workflows. Always follow the selected project's README.md because entry points and artifacts can differ.


Responsible Use

This repository is intended for education, experimentation, technical demonstration, and portfolio presentation.

General limitations include:

  • model outputs depend on the quality and representativeness of the underlying data;
  • accuracy can hide minority-class and class-specific weaknesses;
  • probability scores may not be calibrated unless explicitly validated;
  • synthetic or sample data may not represent real operational populations;
  • neural networks can learn and reproduce bias present in training data;
  • predictions can become unreliable on out-of-distribution inputs;
  • healthcare and financial applications are demonstrations, not diagnostic or approval systems;
  • optimization recommendations depend on model assumptions and defined constraints;
  • portfolio models are not automatically production-ready;
  • no application should be the sole basis for consequential decisions.

Important outputs should be verified through trusted data, domain expertise, independent validation, and appropriate human oversight.


Technical Coverage

Area Demonstrated Through
Binary classification Churn, fraud, credit risk, diabetes, and tabular embeddings
Multi-class classification Handwritten digit recognition
Regression House-price prediction and customer-value forecasting
Multi-task and multi-output learning Customer lifetime value and multi-output prediction
Imbalanced learning Fraud detection and risk-oriented classification
Categorical embeddings Customer lifetime value and tabular deep learning
Computer vision Handwritten digit recognition
Optimization Dynamic pricing
Manual and batch inference Streamlit forms, CSV scoring, and downloadable outputs
Deployment Ten Streamlit Community Cloud applications
Testing and CI/CD Automated tests and project-specific GitHub Actions workflows

Core Skills Demonstrated

Artificial Neural Networks · TensorFlow · Keras · Keras Functional API · scikit-learn · pandas · Feature Engineering · Categorical Embeddings · Classification · Regression · Forecasting · Computer Vision · Multi-task Learning · Multi-output Learning · Class-Imbalance Handling · Threshold Analysis · Probability Scoring · Model Evaluation · Streamlit · Testing · GitHub Actions · CI/CD · Business Translation · Responsible AI Communication


Portfolio Positioning

One-line description: Ten end-to-end artificial neural network projects spanning classification, imbalanced learning, probability scoring, forecasting, optimization, computer vision, multi-task learning, multi-output prediction, and categorical embeddings.

Pinned repository description: Professional ANN and deep-learning portfolio featuring ten deployed projects across churn, fraud, credit risk, customer value, healthcare screening, dynamic pricing, computer vision, regression, multi-output learning, and tabular embeddings—with reusable inference, task-appropriate evaluation, Streamlit applications, testing, and CI/CD.

This portfolio connects naturally to a Quality Data Scientist background through non-confidential quality classification, risk detection, operational forecasting, probability-based prioritization, structured decision support, automated analytical pipelines, reusable model validation, and business-facing prediction applications.


License and Third-Party Materials

The original source code and original documentation in this repository are licensed under the MIT License.

Datasets, pretrained models, model weights, images, and other third-party assets used by the individual projects are not relicensed by this repository. They remain subject to the licenses, terms of use, attribution requirements, and usage restrictions established by their respective owners.

Before reusing any third-party material, review the relevant project documentation, original dataset source, model card, and provider terms. Where applicable, dataset and pretrained-model sources are identified within the corresponding project directory.

Unless explicitly stated otherwise, trained models, evaluation outputs, and generated artifacts in this portfolio are provided for educational, research, and portfolio-demonstration purposes. They are not guaranteed to be suitable for production, medical, financial, safety-critical, or other high-risk applications.


Author

Anmol Tripathi
Quality Data Scientist | Data Science | Machine Learning | Applied AI | Deep Learning | Analytics Engineering | Quality Analytics

About

Ten end-to-end ANN projects covering classification, regression, fraud detection, risk scoring, customer value, optimization, computer vision, multi-output learning, embeddings, and Streamlit deployment.

Topics

Resources

Stars

0 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages