This repository showcases my journey through different domains, applying various machine learning models and techniques to solve real-world problems. Each project is documented, and includes both the code and a comprehensive analysis.
A machine learning project focused on predicting the occurrence of brain strokes. The project applies Logistic Regression, K-Nearest Neighbors, and Random Forest models, including exploratory data analysis (EDA), feature engineering, model training, evaluation, comparison, and fine-tuning. Models used - Logistic Regression, K-Nearest Neighbors, Random Forest.
This project investigates the factors affecting vehicle CO2 emissions in Canada and constructs a Random Forest regression model to predict the emissions based on various features. It includes feature engineering, EDA, visualization, hypothesis testing, model building, and fine-tuning. Model used: Random Forest.
Utilizing Long Short-Term Memory (LSTM) networks implemented with Keras, this project aims to predict Apple Inc.'s stock prices. The project encompasses data preprocessing, model building and training, evaluation, hyperparameter tuning, visualization, and model persistence. Model used: LSTM model implemented with Keras.
A text classification project to identify spam emails. It explores different models, including TF-IDF Vectorization with Multinomial Naive Bayes, Support Vector Machine (SVM), and LSTM networks using TensorFlow and Keras, covering data preprocessing, visualization, model training, evaluation, fine-tuning, and testing on custom data. Models used: TF-IDF Vectorization + Multinomial Naive Bayes, SVM, TensorFlow and Keras:LSTM.
Replicating and extending the work of Franck Dernoncourt and Ji Young Lee's 2017 study on sequential sentence classification in medical abstracts. It leverages the 'PubMed 200k RCT' dataset to explore and evaluate different natural language processing (NLP) models for structuring abstracts into coherent segments. Inspired by methodologies from and guided by the Zero to Mastery TensorFlow course, particularly the SkimLit project. The best model: TensorFlow model with character, token and positional embedding layers.
A deep learning project utilizing Transfer Learning with TensorFlow to classify dog breeds. The model, based on mobilenet_v2_130_224 from TensorFlow Hub and trained on the ImageNet database, demonstrates the power of transfer learning in image classification tasks. Model used - mobilenet_v2_130_224 from TensorFlow Hub.
Note: The project implements methods used in TensorFlow 2.2. Model works, however, some parts of code might seem outdated.
The project was build during completing the bootcamp. The bootcamp's project can be found here
This project leverages the power of advanced machine learning models to classify movie plots into one or more genres (multi-label classification). Two primary models have been utilized and optimized for this task: LSTM (Long Short-Term Memory) and BERT (Bidirectional Encoder Representations from Transformers). I also tried to use data modification to handle prediction minority classes. However some techniques (translation) were not implemented in the training because of high time complexity. This project showed me that it is impossible to always achieve very high results and sometimes experiments do not lead to the expected outcome. I could not achieve high accuracy using the described models and techniques. However it was a useful experience in working with text data. The data is taken from Kaggle.
The projects use various libraries and frameworks, primarily:
- TensorFlow (including Keras)
- scikit-learn
- pandas
- numpy
- matplotlib and seaborn for visualization
For any inquiries or collaboration requests, feel free to contact me or open an issue.
These projects are open-sourced under the MIT license.