Handwriting Recognition using Deep Learning

docs/DL/projects/hand_writing_recognition.md

@@ -0,0 +1,200 @@
+# Handwriting Recognition on MNIST Dataset 
+
+### AIM 
+To develop a system that recognizes handwritten digits using the MNIST dataset, leveraging deep learning techniques for high accuracy.
+
+### DATASET LINK 
+[https://www.kaggle.com/datasets/hojjatk/mnist-dataset](https://www.kaggle.com/datasets/hojjatk/mnist-dataset)
+
+### NOTEBOOK LINK 
+
+[https://colab.research.google.com/drive/1IPZ8Yy0UG_5NdcbwlgHhwMu4vr5fBaqu?usp=sharing](https://colab.research.google.com/drive/1IPZ8Yy0UG_5NdcbwlgHhwMu4vr5fBaqu?usp=sharing)
+
+### LIBRARIES NEEDED 
+
+??? quote "LIBRARIES USED"
+
+    - tensorflow
+    - pandas
+    - numpy
+    - matplotlib
+    - idx2numpy
+    - seaborn 
+    - sklearn
+
+--- 
+
+### DESCRIPTION 
+
+!!! info "What is the requirement of the project?"
+    - The project aims to create a system capable of recognizing handwritten digits.  
+    - This involves in exploring deep learning models for image classification tasks.  
+    - It provides accurate and efficient digit recognition. 
+
+
+??? info "Why is it necessary?"
+    - It help in automating data entry and reduces human errors in digit recognition.  
+    - It forms the foundation of several real-world applications like check reading, postal code recognition, and document scanning.  
+    - It also demonstrates the potential of neural networks for image-based tasks.  
+
+??? info "How is it beneficial and used?"
+    - **Banking sector :** it is used in the banking sector for processing checks.  
+    - **Postal services :** it helps postal services for reading and sorting addresses.  
+    - **Character recognition :** it enables advancements in Optical Character Recognition (OCR) technologies. 
+
+??? info "How did you start approaching this project? (Initial thoughts and planning)"
+    - **Data Exploration :** Ive made attempts to understand and explore MNIST dataset for understanding the data structure.  
+    - **CNN :** Ive did some research on Convolutional Neural Networks (CNNs) for image classification.  
+    - **Optimizers :** Experimented with different architectures to optimize accuracy such as SGD[Stochastic Gradient Descent] and Adam[Adaptive Moment Estimation]. 
+
+??? info "Mention any additional resources used (blogs, books, chapters, articles, research papers, etc.)"
+    - [https://www.mygreatlearning.com/blog/how-to-recognise-handwriting-with-machine-learning/](https://www.mygreatlearning.com/blog/how-to-recognise-handwriting-with-machine-learning/)
+    - [https://cs231n.stanford.edu/reports/2017/pdfs/810.pdf](https://cs231n.stanford.edu/reports/2017/pdfs/810.pdf)
+    - [https://youtu.be/iqQgED9vV7k?si=PRcpO-Fx4HeqFXuz](https://youtu.be/iqQgED9vV7k?si=PRcpO-Fx4HeqFXuz)
+
+--- 
+
+### EXPLANATION
+
+#### DETAILS OF THE DIFFERENT FEATURES 
+
+ - **Dataset**: MNIST dataset contains 60,000 training and 10,000 testing images of handwritten digits (0-9).  
+ - **Preprocessing**: Normalized pixel values and reshaped images for compatibility with CNNs.  
+ - **Model Architecture**: Built using TensorFlow/Keras, consisting of convolutional layers, max pooling, and fully connected layers.  
+ - **Evaluation Metrics**: Accuracy and loss tracked during training and testing phases. 
+
+--- 
+
+#### WHAT I HAVE DONE 
+
+=== "Step 1"
+
+    Initial data exploration and understanding:
+
+      - Loaded the MNIST dataset using TensorFlow/Keras.
+      - Explored the structure of the dataset, including image dimensions and class distribution.
+      - Visualized sample images to understand the nature of handwritten digits.
+
+=== "Step 2"
+
+    Data cleaning and preprocessing:
+
+      - Normalized pixel values to a range of [0, 1] for better convergence during training.
+      - Reshaped the images to match the input shape required by the CNN.
+      - Split the dataset into training, validation, and testing sets.
+
+=== "Step 3"
+
+    Feature engineering and selection:
+
+      - No manual feature engineering required as the CNN extracts features automatically.
+      - Applied data augmentation techniques (e.g., rotation, zoom, and flipping) to expand the dataset size and improve generalization.
+
+=== "Step 4"
+
+    Model training and evaluation:
+
+      - Built a Convolutional Neural Network (CNN) architecture with convolutional, max-pooling, dropout, and fully connected layers.
+      - Used the training set to train the CNN and the validation set to monitor performance and prevent overfitting.
+      - Evaluated the model using metrics such as accuracy and loss on both training and testing sets.
+
+=== "Step 5"
+
+    Model optimization and fine-tuning:
+
+      - Tuned hyperparameters such as learning rate, batch size, and number of epochs.
+      - Applied dropout to reduce overfitting and improve generalization.
+      - Experimented with different CNN architectures to achieve better accuracy.
+
+=== "Step 6"
+
+    Validation and testing:
+
+      - Validated the model on a separate validation set to ensure it generalizes well to unseen data.
+      - Tested the final trained model on the test dataset to assess real-world performance.
+
+--- 
+
+#### PROJECT TRADE-OFFS AND SOLUTIONS 
+
+=== "Trade-Off 1"  
+
+    **Training time vs. model complexity :**  
+
+    - Used a simpler architecture to balance accuracy and computational cost.  
+
+=== "Trade-Off 2"  
+
+    **Model accuracy vs. generalization :**  
+
+    - Regularized the model using dropout layers to prevent overfitting. 
+
+--- 
+
+### SCREENSHOTS 
+
+!!! success "Project structure or tree diagram"
+
+    ``` mermaid  
+    graph LR  
+        A[Load MNIST Dataset] --> B[Preprocess Data];  
+        B --> C[Build CNN Model];  
+        C --> D[Train Model];  
+        D --> E[Evaluate Model];  
+        E --> F[Save and Deploy Model];  
+    ```  
+
+??? tip "Visualizations and EDA of different features"
+
+    === "Converted Images"
+        ![Converted Images](https://github.com/user-attachments/assets/4934eb8e-6644-43ce-989a-1094b3a5105d)
+
+    === "Prediction"
+        ![Prediction](https://github.com/user-attachments/assets/549da012-a1a9-4e43-aa4b-a55057258259)
+
+??? example "Model performance graphs"
+
+    === "Confusion Matrix"
+        ![confusion matrix](https://github.com/user-attachments/assets/7676ac47-eb62-4a24-9d80-6a44d5d2f9e7)
+
+--- 
+
+### MODELS USED AND THEIR EVALUATION METRICS 
+
+| Model | Accuracy | MSE | R2 Score |
+|-------|----------|-----|-------|
+| CNN | 95.34% | - | - |
+
+--- 
+
+### CONCLUSION 
+
+#### KEY LEARNINGS 
+
+!!! tip "Insights gained from the data"
+    - **Data Variety :** The importance of preprocessing for enhancing model performance.  
+    - **Feature Extraction :** How CNNs extract features and learn patterns in images. 
+
+??? tip "Improvements in understanding machine learning concepts"
+    - **Model Selection :** Strengthened knowledge of convolutional layers and pooling mechanisms. 
+    - **TensorFlow :** Enhanced Knowledge of Tensorflow/Keras modules.
+
+??? tip "Challenges faced and how they were overcome"
+    - **Overfitting :** Managed overfitting by applying dropout and data augmentation. 
+
+--- 
+
+#### USE CASES 
+
+=== "Application 1"  
+
+    **Banking Sector**  
+
+      - Automatic digit recognition for processing checks.  
+
+=== "Application 2"  
+
+    **Postal Services**  
+
+      - Automating the reading of zip codes for faster mail sorting.  
+