Karsten_Portofolio

Karsten Errando Winoto — Portfolio

📌 Projects

• Project 1: HeartGuard
• Project 2: Transformer Leopard Gecko
• Project 3: Cooperative App
• Project 4: ECG Heart Disease Classification

---

Project 1: HeartGuard

This is a project I developed for my Mobile Application Programming course at university, where I built a mobile application that integrates a machine learning model for heart disease classification.

• Dataset: Kaggle Heart Disease Dataset
• Model: Random Forest
• Frontend: Flutter framework
• Backend: Local API built with Flask (integrated with the ML model)

📱 Mobile App Layout

   

📊 Model Accuracy

---

Project 2: Transformer Leopard Gecko

This is a project I completed for my Pattern Recognition course as the final project. The goal was to classify Leopard Gecko morphs using a Transformer-based model.

• Classes:
  • Boldstripe
  • Albino
  • Albino Boldstripe
• Dataset: ~60 images collected manually (Google, Instagram, etc.)
• Architecture: Transformer
• Training: 100 epochs
• Achieved test accuracy: 41%

🗂️ Dataset Samples

• Boldstripe
  

• Albino
  

• Albino Boldstripe
  

🧪 Evaluation Results

• Test Accuracy
  

• Train Accuracy and Loss
  

---

Project 3: Cooperative App

This is a mobile cooperative application I built during my internship at Dua Kelinci.
It allows factory employees to browse products, manage shopping carts, perform checkouts, and view transaction history.

🔧 Tech Stack

• Frontend: Flutter (Android/iOS)
• Backend: Flask (RESTful API)
• Database: MySQL

⚙️ Core Features

• Product catalog browsing with detailed views
• Add to cart, update quantities, and remove items
• Checkout flow with order summary and payment-ready integration
• Transaction history with order status tracking
• User authentication (login/logout, session management)

🏗️ Architecture

• RESTful API built with Flask
• Persistent storage in MySQL
• Frontend–backend communication via JSON over HTTPS

---

Project 4: ECG Heart Disease Classification with Deep Learning

This is my Final Thesis project, where I developed a classification model for heart disease based on ECG signal amplitudes using multiple deep learning methods.

📊 Dataset

• Source: Kaggle - ECG Heartbeat Categorization Dataset
• Training set: 87,554 samples
• Testing set: 21,892 samples
• Each ECG signal length: 1870 ms (120 Hz sampling rate)

Class Distribution (before preprocessing)

Class	Training	Testing
Normal	72,471	18,118
Fusion of paced and normal	6,431	1,608
Premature ventricular contraction	5,788	1,448
Atrial Premature	2,223	556
Fusion of ventricular and normal	641	162
Total	87,554	21,892

Class Distribution (after splitting)

Class	Training	Validation	Testing
Normal	57,892	14,579	18,118
Fusion of paced and normal	5,182	1,249	1,608
Premature ventricular contraction	4,676	1,112	1,448
Atrial Premature	1,797	426	556
Fusion of ventricular and normal	496	145	162
Total	70,043	17,511	21,892

Signal Samples

• Normal
  

• Atrial Premature
  

• Premature Ventricular Contraction
  

• Fusion of Ventricular and Normal
  

• Fusion of Paced and Normal
  

🏗️ Models Used

• CNN1D (Convolutional Neural Network)
  

• RNN (Recurrent Neural Network)
  

• LSTM (Long Short-Term Memory)
  

• GRU (Gated Recurrent Unit)
  

📈 Model Evaluation

Training Performance

• CNN
  

• RNN
  

• LSTM
  

• GRU
  

Testing Performance

• CNN
  

• RNN
  

• LSTM
  

• GRU
  

Final Results

Model	Accuracy	Precision	Recall	F1-Score
CNN	0.98	0.92	0.90	0.91
RNN	0.83	0.17	0.20	0.18
LSTM	0.97	0.89	0.79	0.82
GRU	0.98	0.90	0.89	0.90

Key Insight: CNN and GRU achieved the best performance with 98% accuracy, while RNN underperformed due to underfitting.

---