A collection of jupyter notebooks on ML and DL projects. Datasets come mostly from Kaggle (credits in the notebooks).
It is recommended to open the notebooks in Google Colab, to correctly display all interactive plots.
A comparison between classical bag-of-words and DL models (LSTM recurrent neural network and RoBERTa attention-based transformer) to classify the polarity of english Amazon reviews. The accuracy of the best model is over 96%, with consistent precision and recall.
We build a residual neural network which analyzes brain xray scans of patients and detects the presence of brain tumors. Trained on around 4000 images, the model reaches an accuracy of over 99%. We also provide a GUI widget for quick inference, which allows the user to upload the image of a brain scan and get the model prediction.
...and more on Kaggle:
In-depth EDA of a dataset of 20 features related to bank customers with a credit card. We compare different tree-based learning models, like Random Forest and XGBoost, and neural networks to build a classifier which predicts whether a customer will churn. The best model has an AUC score of 99% and detects churning customers with a 97% recall.