Md. Kabin Hasan Kanchon

Education

Bachelor of Science (2019-2023)
North South University
Computer Science Engineering
CGPA : 3.59
Major in Artificial Intelligence & Minor in Networks

Higher School Certificate (2018)
Savar Cantonment Public School & College
Science
GPA : 4.50

Secondary School Certificate (2016)
Radio Colony Model School
Science
GPA : 5.00

Projects

1. Trending FIFA players
   This is a trending FIFA player's data analysis project where we have done web scraping using selenium from this website.
   Click here for detail..
   
   Github link: trending_FIFA_players
   Tableau dashboard link: Data visualization in Tableau
   Kaggle link: Trending FIFA Players Dataset
   


2. Sports Ball Recognizer
   An image classification model from data collection, cleaning, model training, deployment and API integration.
   Click here for detail..
   
   Github link: Sports-Ball-Recognizer
   GitHub Pages link: Sports Ball Recognizer
   Kaggle link: Sports Ball Dataset
   


3. Multilabel Quote Classifier
   A text classification model from data collection, model training, and deployment.
   Click here for detail..
   
   Github link: Multilabel-Quote-Classifier
   Web page link: Multilabel Quote Classifier
   Kaggle link: Multilabel Quote Classifier
   

Multilabel Quote Classifier

A text classification model from data collection, model training, and deployment. The model can classify 675 different types of quote tags

Data Collection

1. The quotes URLs were scraped and stored along with the quote title
2. Using the URLs, the quotes, the authors, the category and the description URLs are scraped and stored
3. The final part was tag scraping and it was the difficult one. I split the total data into mini-batches and scrap tags and stored

Model Training

Finetuned distilroberta-base and bert-base-uncased models from HuggingFace Transformers using Fastai and Blurr. The model training notebook can be viewed here

Result Comparison

Model Compression and ONNX Inference

The trained 'Bert Base Uncased' model has a memory of 422+MB. I compressed this model using ONNX quantization and brought it under 106MB.

Model Deployment

Sports Ball Recognizer

An image classification model from data collection, cleaning, model training, deployment and API integration. The model can classify 10 different types of balls The types are the following:

• Football
• Basketball
• Volleyball
• Rugby
• Golf
• Cricket
• Tennis
• Bowling
• Billiards
• Baseball

Dataset Preparation

Data Collection: Downloaded from DuckDuckGo using term name and adding extra string ' only balls images'. A total of 2872 images were collected initially. After unlinking the damaged data, the total number of images was 2776.
DataLoader: Used FastAI DataBlock API to set up the DataLoader.
Data Augmentation: FastAI provides default data augmentation which operates in GPU.
For the data preparation code click here

Training and Data Cleaning

Deployment

Hugging Face: I deployed the model to the HuggingFace Spaces Gradio App. The implementation can be found here
GitHub pages: The deployed model API is integrated here

Trending FIFA Players

This is a trending FIFA player's data analysis project where we have done web scraping using selenium from this website. We scrapped 3000 trending FIFA players' data and stored them in an Excel sheet. There are 12 columns as given:

• Player_name
• Images
• Age
• National_team
• Positions
• Overall
• Potential_overall
• Current_club
• Current_contract
• Value
• Wage
• Total_stats

Problem statements

Then we have done some data processing using Python libraries like pandas and re. By organising the data in Tableau, we tried to understand:

1. Which country has how many players on the trending FIFA players website and what is the total value?
2. Who are the top players in the trending sheet?
3. Players of which age have good stats?
4. Which players are ruling in the trending player's list?
5. Top clubs having the most valued players in the trending list?

Findings and Observations from the Dashboard

Player Performance Analysis in clubs:

• From the bar chart, we can see the top 20 players according to their overalls. They are in different colours in respect of their current clubs.
• From the pie chart, we find out that Real Madrid has the most valuable players. And also the other clubs and their values that are on the top list.
• From the density graph, we can see that approximately players between 20-33 years old have better stats than others.
• The bubble graph shows the top 20 young players (18-25 years old) and their value comparison. They are also coloured according to their current clubs.

Player Performance Analysis in national teams:

• From the first bar chart, we can see that Spain has the most value and England has the most number of players. The bar chart shows the national team rankings by the total value in the trending list. It also shows the total number of players on the list.
• From the graphical graph, we can find out each country's values and total number of players by their spatial position.
• From the last bar chart, we can see the same top player rankings but coloured concerning national teams.

Researches

1. Learning Style Detection and Content Modifier with Artificial Intelligence
   The project aims to use Artificial Intelligence (AI) to identify individual learning styles and adapt educational content accordingly in the rapidly advancing field of EdTech. It employs activity and feedback classification techniques, such as analyzing web tracking logs and individual responses, to categorize learners' styles. Using clustering and text classification methods, the project achieved 100% accuracy in classifying learners into Visual, Auditory, and Kinesthetic styles. Content modifications, including color-coding, audio, mind-maps, and flashcards, are tailored to suit these identified learning styles. Overall, the project emphasizes personalized learning experiences through AI-driven content adaptation in EdTech.

   Too see the paper click here.

2. Machine Learning Approach to Smoking Detection by Body Signals
   The study examines smoking's impact on individuals and bystanders, leveraging wearable sensors to objectively track daily smoking habits. Using bio-signals like hemoglobin, cholesterol type, and urine protein, a new machine-learning system identifies smoking presence. It also accounts for factors like breathing rate and blood pressure to distinguish smoking-related changes from other biological activities. Six machine learning methods, including Logistic Regression, Decision Tree, KNN, SVM, Adaboost, and Random Forest, were used to detect smokers. Logistic Regression stood out with 74.34% accuracy, proving most effective among the methods in identifying smokers based on collected bio-signals.

   Too see the paper click here.

3. Comparing Deep Learning Techniques for Imbalanced Dataset
   The study focuses on early detection of Retinal Fundus Multi-Disease and Skin Cancer lesions, crucial for successful treatment. It explores efficient models like ResNet50, VGG16, and EfficientNet using datasets (HAM10000, ISIC, ODIR). Techniques like class weight, focal loss, and their combination address dataset imbalance. The Model Soup Algorithm achieved a high accuracy of 95.81%, showcasing its effectiveness in early disease identification.

   Too see the paper click here.

Experiences

1. Teaching Assistant (Jun 2022 - Jul 2023)
   North South University
   Department of Mathematics and Physics
   Subjects: Linear Algebra & Business Mathematics
   Responsibilities:
   

   • Assisting Teachers with assignments and assesments.
   • Updating marks in Excel sheets
   • Assisting students with their lessons
   • Proctoring in examinations




2. Data Science Trainee (Oct 2023 - Jan 2024)
   MasterCourse
   Dokkho Cohort-3
   Learnings:
   

   • Web scraping using Selenium, Fastbook etc.
   • Data analysing using Tableau
   • implementation of Machine Learning and Deep Learning models
   • Using Fast.Ai
   • Deploying models in different platforms like HuggingFace spaces, GitHub pages etc.
   • Deploying Flask APIs
   • Hosting in Render, Heroku etc.

Certifications

1. Introduction to Cybersecurity (2022)
   CISCO
   Grameenphone Academy
   You can see the certification badge by clicking here.

2. Data Science Cohort-3 (2023-2024)
   Mastercourse
   Dokkho Career Program, Dhaka
   

3. Deutschkurs A1-A2 (2023)
   Language Course
   Goethe Institut, Dhaka
   

Skills

• Artificial Intelligence
  
  1. Proficient in applying Machine Learning & Deep Learning models.
  2. Particularly skilled in implementing Transfer Learning techniques.
  3. Proficiency extends to data analysis using Tableau.
  4. Adeptness in integrating API interfaces within HuggingFace & Git pages.
  5. Experienced in conducting web scraping operations employing Selenium.
  6. Coupled with fluency in Python programming language.
  7. Comfortable with PyTorch with a readiness to acquire proficiency in TensorFlow.
  8. Proficient utilization of Visual Studio, Jupyter Notebook, Google Colab and Pycharm for development tasks.
  9. Well-versed in GitHub's functionalities and practices for version control and collaboration in projects.
• Networking
  
  1. Experienced in leveraging Cisco Packet Tracer for network design, simulation, and configuration
  2. Proficient in employing Wireshark for in-depth network protocol analysis and troubleshooting.
  3. Possess a strong command of IP calculation techniques and adeptness in various methods for network segmentation and subdivision.
• Software and Web Development
  
  1. Skilled in utilizing Android Studio to develop Android applications.
  2. Adept in harnessing the functionalities of Firebase.
  3. Knowledgeable in working with the Laravel framework for web development purposes.
  4. Proficient across various programming languages such as Python, Java, C++, C, and PHP
  5. Well-acquainted with front-end technologies including HTML, Bootstrap, CSS, and XML for crafting engaging web and mobile applications.
• Other Skills
  
  1. Experienced in the creation, simplification, and conversion of Automatas and Context-Free Grammars (CFGs) for computational models.
  2. Proficient in Microsoft Word, Excel, and PowerPoint, as well as adept in utilizing Google Docs, Sheets, and Slides for comprehensive documentation, data analysis, and impactful presentations.
• Soft Skills
  
  1. Computer Software and Hardware Knowledge, proficient in comprehensive understanding and expertise encompassing a broad spectrum of computer software and hardware components, functionalities, and their interplay within computing systems.
  2. Effective Communication, proficient in conveying ideas clearly and articulately, fostering seamless collaboration among teams.
  3. Adaptability, skilled in navigating diverse environments, swiftly adjusting strategies to meet evolving project needs.
  4. Problem-Solving, Adept at analyzing complex issues, identifying solutions, and implementing effective resolutions.
  5. Time Managesment, proficient in prioritizing tasks, optimizing productivity, and meeting deadlines efficiently.
  6. Team Player, valued for the ability to collaborate seamlessly within multidisciplinary teams, contributing positively to group dynamics.
  7. Leadership, capable of guiding and inspiring others, showcasing leadership qualities to drive projects forward.
  8. Attention to Detail, meticulous in ensuring accuracy and precision, maintaining high standards in all tasks undertaken.
  9. Critical Thinking, skilled in evaluating situations from multiple perspectives, enabling innovative and strategic decision-making.
  10. Resilience, exhibits a resilient attitude when faced with challenges, demonstrating persistence and determination in overcoming obstacles.
  11. Empathy, capable of understanding others' perspectives, fostering a supportive and inclusive work environment.

Co-curricular

1. First prize (2022)
   Introduction to Cybersecurity Quiz Competition
   Grameenphone Academy
   

2. General Member (2021)
   Human Resources Club
   North South University
   

3. Participant (---)
   AI seminars & workshops
   North South University
   

Elements

i = 0;

while (!deck.isInOrder()) {
    print 'Iteration ' + i;
    deck.shuffle();
    i++;
}

print 'It took ' + i + ' iterations to sort the deck.';