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.


Customer Transaction Prediction

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


Vis accumsan feugiat adipiscing nisl amet adipiscing accumsan blandit accumsan sapien blandit ac amet faucibus aliquet placerat commodo. Interdum ante aliquet commodo accumsan vis phasellus adipiscing. Ornare a in lacinia. Vestibulum accumsan ac metus massa tempor. Accumsan in lacinia ornare massa amet. Ac interdum ac non praesent. Cubilia lacinia interdum massa faucibus blandit nullam. Accumsan phasellus nunc integer. Accumsan euismod nunc adipiscing lacinia erat ut sit. Arcu amet. Id massa aliquet arcu accumsan lorem amet accumsan.

![](/images/work-book-cover-2.jpg)

Amet nibh adipiscing adipiscing. Commodo ante vis placerat interdum massa massa primis. Tempus condimentum tempus non ac varius cubilia adipiscing placerat lorem turpis at. Aliquet lorem porttitor interdum. Amet lacus. Aliquam lobortis faucibus blandit ac phasellus.

![](/images/work-book-cover-3.jpg)

Vis accumsan feugiat adipiscing nisl amet adipiscing accumsan blandit accumsan sapien blandit ac amet faucibus aliquet placerat commodo. Interdum ante aliquet commodo accumsan vis phasellus adipiscing. Ornare a in lacinia. Vestibulum accumsan ac metus massa tempor. Accumsan in lacinia ornare massa amet. Ac interdum ac non praesent. Cubilia lacinia interdum massa faucibus blandit nullam. Accumsan phasellus nunc integer. Accumsan euismod nunc adipiscing lacinia erat ut sit. Arcu amet. Id massa aliquet arcu accumsan lorem amet accumsan.


Read next

Customer Transaction Prediction

Image Classification_Dog Vs. Cats