Data.  Curiosity.  Learning.  Creativity.

Welcome to my portfolio website!

Albert Einstein

Hi, welcome to my project! Feel free to check out my github for further details!

https://github.com/vvwzy/DogsVsCats

This classic image classification project with 25,000 images of cats and dogs is very interesting to play with.

When doing this project, I was able to get myself more familiar with **TensorFlow** as well as **Keras**. Also, I practiced my skills in terms of general techniques with **Transfer Learning** using **ResNet50** under CNN architecture which has been proven many times to be able to achieve high accuracy in performance.

Also, I have attached a bunch of reflections by doing this project at the end of the code as comment. Feel free to check them out if you are interested in. 😃


Image Classification_Dog Vs. Cats

Hi, welcome to my project! Feel free to check out my github page to learn more

https://github.com/vvwzy/SantanderCustomerTransactionPrediction

Throughout this project, I was trying to find out if there is a way to determine which customers will make a specific transaction in the future, irrespective of the amount of money transacted.

The Python Libraries that have been used for this project including **Numpy, Pandas, sklearn, scipy, skopt, matplotlib, xgboost, lightgbm, tensorflow and bayes_opt.**

After understanding the business background and the problem statement, I started exploring the dataset doing **basic EDA**. There were 200k observations and 200 features in the training dataset and same size for the testing data. Considering all these 200 features, I needed to find the most relevant features that help to make better predictions.

The next step was **feature engineering** since the features that went into the model are crucial for training a good prediction model. Then I checked for missing values and repeated values and found that there wass none of them.

After the above, I started to try different models including **Logistic Regression, Random Forest, XGBoost and LightGBM**. After comparing these models, I decided to stick with **LightGBM**. Finally after all algorithms have been validated, trained and tested, I ended up using LightGBM model with the help of **Bayesian Optimization** to **pick the best hyperparameters** and **Stratified K Fold** to elevate the **AUC score** on our validation dataset. After the model being built, by plotting the feature importance among all the features I had, I knew the weighted contribution of each feature to final predictions.


Welcome to my portfolio website!

Take a look at my projects :)

Image Classification_Dog Vs. Cats

Customer Transaction Prediction

Home

Welcome to my portfolio website!

Take a look at my projects :)

Image Classification_Dog Vs. Cats

Customer Transaction Prediction