• 👨‍💻 Data Scientist by profession, leveraging data to drive insights and solutions.
• 📜 Holds a Bachelors Degree in Computer Engineering from Mapúa University 🐦,
• 🌟 Passionate about all things programming and technology.
• 🧠 Currently immersed in the fascinating world of Machine Learning and Deep Learning.
• 👨‍💻 My hobby is coding, which became my job and now I don't have any other hobby 😅

This section is currently under construction, and the source codes have not yet been transferred to Git. If you require further information, please don't hesitate to reach out to me. Thank you for your understanding. ❤️

Keywords: Deep Learning | Computer Vision | Convolutional Neural Network (CNN) | HAM10000 | Pre-Augmentation | Transfer learning | EfficientNetV2 | Soft-Attention | View Source

• Skin cancer is the most widespread form of cancer, with melanoma being its deadliest variant, responsible for 75% of skin cancer-related deaths. Early detection is paramount for effective treatment and positive outcomes.

• In this project, I developed a Deep Convolutional Neural Network (CNN) model capable of classifying 7 types of skin lesions, achieving an 87% accuracy and demonstrating strong class discrimination with an 0.97 AUC.

• The HAM10000 dataset by Tschandl, P., Rosendahl, C., & Kittler, H., consisting of 10,015 dermoscopic images, was used for model training and validation. Downsampling and Pre-Augmentation were performed to address the dataset's significant imbalance.

• Transfer learning techniques were employed to achieve optimal performance while minimizing training time. EfficientNetV2 was the chosen pre-trained CNN architecture due to striking the best balance between model performance and hardware usage, as determined through testing.

• Additionally, the concept of Soft-Attention, based on the study of Datta et al., was implemented to visualize the area of focus of the model. when identifying its class.

Keywords: Deep Learning | Natural Language Processing | Text Classification | Recurrent Neural Network | Long Short-Term Memory | Spam Reviews | Batch Processing | View Source

• In today's world, online shopping has become incredibly common, reshaping global commerce and expected to generate $3.2 trillion in revenue by 2024. 93% of consumers states that their purchasing decisions are heavily influenced by reviews, there is a clear need for a system capable of detecting false or spam reviews to safeguard genuine feedback and enable informed choices.

• A Text Classification model was constructed using Natural Language Processing (NLP) techniques. The model is based on a Recurrent Neural Network (RNN) model, specifically utilizing Long Short-Term Memory (LSTM) cells. It attains an accuracy of 89% in determining whether a review is spam or genuine.

• I utilized the Amazon Product Review (Spam and Non-Spam) dataset from Naveed Hussain et al., with a total size of 18.4GB comprising of 26.7 million reviews distributed across six product categories. Each category is represented by a JSON file containing respective reviews.

• Due to the RAM memory constraints and the dataset's JSON format, extensive preparation was necessary. I focused on reviews for fashion and clothing products, which ranked second in sales distribution on Amazon in 2022, representing 24.7% of total sales, with a JSON file size of 3.21GB. The dataset was then converted into a CSV file format containing 5.7 million reviews through batch processing, preventing memory overload and facilitating further processing and analysis.

• Image Classification: Laboratory Apparatus Indetification
• Image Classification: One-Class Convolutional Neural Network based on a study
• Regression Model: Algae Count Prediction
• Clustering Model: Customer Segmentation using both Numerical and Categorical Data
• NLP Sentiment Analysis: Twitter Comment Polarity Prediction