https://github.com/ayushpatel2026/cnn-malaria
End-to-end ML Project - CNN Classifier
https://github.com/ayushpatel2026/cnn-malaria
dvc machine-learning mlops python tensorflow
Last synced: about 2 months ago
JSON representation
End-to-end ML Project - CNN Classifier
- Host: GitHub
- URL: https://github.com/ayushpatel2026/cnn-malaria
- Owner: Ayushpatel2026
- Created: 2024-11-28T00:05:57.000Z (over 1 year ago)
- Default Branch: main
- Last Pushed: 2024-12-16T13:56:15.000Z (over 1 year ago)
- Last Synced: 2025-01-30T01:14:42.830Z (over 1 year ago)
- Topics: dvc, machine-learning, mlops, python, tensorflow
- Language: Jupyter Notebook
- Homepage:
- Size: 97.2 MB
- Stars: 0
- Watchers: 1
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
# **Malaria Diagnosis with Convolutional Neural Networks**
This project leverages Convolutional Neural Networks (CNNs) to classify blood cell images as either infected (parasitized) or uninfected with malaria. Achieving an accuracy of **94%**, this solution demonstrates the use of modern machine learning techniques and MLOps principles to build a robust, modular, and collaborative pipeline.
## **Key Features**
- **High Accuracy**: Achieved 94% classification accuracy on the dataset.
- **End-to-End Pipeline**: Built modular pipelines for:
- **Data Ingestion**: Loading and managing the dataset.
- **Data Processing**: Preprocessing images and applying data augmentation.
- **Model Training**: Custom training routines with learning rate scheduling and regularization.
- **Model Evaluation**: Evaluating model performance using test data.
- **MLOps Principles**: Designed for **reproducibility** and **collaboration**:
- Modular and reusable code.
- Streamlined pipelines for clear separation of tasks and used DVC to keep track of the pipelines
- Scalable and extensible for future improvements.
- **Technologies Used**:
- **TensorFlow** and **Keras**: Model development and training.
- **Matplotlib**: Data visualization and analysis.
- **Flask**: To make a user friendly web application
- **DVC**: To keep track of the pipelines and models for reproducibility and traceability
- **Jupyter Notebook**: For initial prototyping and testing.
## **Project Workflow**
### 1. **Initial Prototyping**
- Explored and visualized the dataset and tested early CNN models in **Jupyter Notebook** (research/MalariaDiagnosis.ipynb).
- Fine-tuned hyperparameters, implemented **data augmentation**, and evaluated initial performance.
### 2. **End-to-End Pipeline**
- Transitioned to a **modular pipeline structure** for scalability and collaboration:
- **Data Ingestion**: Reads and prepares data for training and testing.
- **Data Processing**: Applies transformations like resizing, normalization, and augmentation.
- **Training**: Trains the CNN using the ADAM optimizer and learning rate scheduling for better performance.
- **Evaluation**: Validates the model's accuracy and loss on test data.
### 3. **Model Features**
- **Data Augmentation**: Increases dataset diversity with techniques like flipping and rotation.
- **Learning Rate Scheduling**: Dynamically adjusts the learning rate during training to optimize convergence.
- **Custom Callbacks**: Monitors performance metrics and enables early stopping to prevent overfitting.
# Flask App
A wep application was made using Flask to provide a user friendly way to upload cell images and get a prediction from the model.
Below are two images from the wep application: with a cell that is uninfected and one with a parasitized cell.
## **Example: Uninfected Cell**
## **Example: Parasitized Cell**
# How to run?
### STEPS:
Clone the repository
```bash
git clone https://github.com/Ayushpatel2026/cnn-malaria
```
### STEP 01- Create a conda environment after opening the repository
```bash
conda create -n cnn-malaria python=3.9 -y
```
```bash
conda activate cnn-malaria
```
### STEP 02- install the requirements
```bash
pip install -r requirements.txt
```
```bash
# Finally run the following command
python app.py
```
Now, open up the application on your local host and port