🧠 Deep Learning Projects by Govind Bhatter

---

📚 Table of Contents

• 🚀 Project 1: DDoS Attack Classification
• 🌌 Project 2: Stellar Classification
• 🏞️ Project 3: Intel Image Classification
• 🌿 Project 4: Cassava Leaf Disease Classification
• 🧰 Tech Stack

---

🚀 Project 1: DDoS Attack Classification

🔍 Overview

This project focuses on detecting and classifying Distributed Denial of Service (DDoS) attacks using a deep learning model. DDoS attacks aim to disrupt the normal traffic of targeted systems by overwhelming them with traffic. Early and accurate classification is vital to defend against such cyber threats.

📊 Dataset Description

The dataset contains network traffic logs with both normal and malicious (DDoS) data. Key features include:

• Source IP, Destination IP
• Source Port, Destination Port
• Protocol type, Packet Length, Packets/Time
• Highest Layer, Transport Layer
• target: Binary label indicating DDoS (1) or normal (0)

⚙️ Methodology

1. Preprocessing

   • One-hot encoding for categorical features
   • Feature scaling with StandardScaler
   • Train-test split for supervised learning

2. Model Architecture

   • Multi-layered dense neural network
   • Activation: ReLU (hidden), Sigmoid (output)
   • Loss function: binary_crossentropy
   • Optimizer: Adam
   • Metrics: Accuracy, Precision, Recall

3. Evaluation

   • Validation using test split
   • Confusion matrix and loss/accuracy plots for analysis

🎯 Objective

To create a robust classifier that can differentiate between benign and malicious traffic in real-time network environments.

👉 View Project on Kaggle

---

🌌 Project 2: Stellar Classification

🔍 Overview

This project classifies stars based on their spectral and photometric features using deep learning. Stellar classification helps astronomers understand the physical properties and life cycle of stars.

📊 Dataset Description

The dataset includes data from astronomical observations with key features:

• u, g, r, i, z: Magnitudes from photometric filters
• alpha, delta: Right Ascension and Declination
• redshift: Redshift value
• class: Target variable (GALAXY, STAR, or QSO)

Identifiers like obj_ID, spec_obj_ID, run_ID, and plate were excluded during training.

⚙️ Methodology

1. Preprocessing

   • Removed irrelevant identifiers
   • Encoded target classes
   • Normalized features
   • Train-test split

2. Model Architecture

   • Dense layers with ReLU activation
   • Output layer with Softmax for 3-class classification
   • Loss function: sparse_categorical_crossentropy
   • Optimizer: Adam

3. Evaluation

   • Classification report with accuracy and confusion matrix
   • Learning curves and performance heatmaps

🎯 Objective

To predict the type of astronomical object (star, galaxy, or quasar) using its photometric and positional attributes.

👉 View Project on Kaggle

---

🏞️ Project 3: Intel Image Classification

🔍 Overview

A CNN-based image classifier trained on natural scene images across 6 categories using the Intel Image Classification dataset.

📊 Dataset Description

Images of size 150x150 pixels across 6 categories:

• buildings, forest, glacier, mountain, sea, street

Includes separate training, validation, and test sets.

⚙️ Methodology

1. Preprocessing

   • Resizing and normalization of image data
   • Augmentation using ImageDataGenerator
   • Batch loading for training/validation

2. Model Architecture

   • Convolutional layers with ReLU and MaxPooling
   • Fully connected layers with Dropout
   • Output layer with Softmax for 6-class prediction
   • Loss function: categorical_crossentropy
   • Optimizer: Adam

3. Evaluation

   • Accuracy and loss on validation/test sets
   • Classification report and sample prediction outputs

🎯 Objective

To train a deep learning model capable of accurately classifying natural scenes into one of the six predefined categories.

👉 View Project on Kaggle

---

🌿 Project 4: Cassava Leaf Disease Classification

🔍 Overview

This project focuses on detecting and classifying cassava leaf diseases using deep learning. Early detection of diseases like CBB, CBSD, CGM, CMD is vital to prevent crop loss and improve productivity. The model uses EfficientNetB3 for transfer learning after experimenting with self-made architectures and other pre-trained models.

📊 Dataset Description

The dataset contains images of cassava leaves categorized into five classes:

• 0 – Cassava Bacterial Blight (CBB)
• 1 – Cassava Brown Streak Disease (CBSD)
• 2 – Cassava Green Mottle (CGM)
• 3 – Cassava Mosaic Disease (CMD)
• 4 – Healthy Leaf

Images are resized and augmented for robust training.

⚙️ Methodology

1. Preprocessing

   • Resizing to 300×300 pixels
   • Data augmentation: horizontal/vertical flips, random rotations

2. Model Architecture

   • Transfer learning using EfficientNetB3
   • Top layers: GlobalAveragePooling, Dense layers, Dropout
   • Output layer: Softmax for 5-class classification

3. Training

   • Optimizer: Adam
   • Loss: Categorical Crossentropy
   • Metrics: Accuracy, Precision, Recall

4. Evaluation

   • Accuracy, Loss, Precision, Recall, F1-score
   • Visualizations for training vs validation performance

🎯 Objective

To develop a reliable deep learning model capable of automatically classifying cassava leaf images into healthy or diseased categories.

👉 View Project on Kaggle

---

🧰 Tech Stack

---

✨ Feel free to ⭐ this repo if you find these projects helpful!