Projects
Donor Choose Classification
-
100k project proposals were classified into two classes using Decision Tree, Naive Bayes, XGBoost and LSTM and cross-validated
-
Pre-processed data, including encoding with TF-IDF and Word2Vec, visualizing with t-SNE, transforming and performing initial analysis
-
Conducted hyper-parameter tuning using k-fold cross validation to get the maximum AUC score.
Microsoft Malware Detection
-
Used Random Forest and XGBoost classifiers to identify malware using .byte and .asm files
-
Pre-processed data, including cleaning, encoding, visualizing, transforming and performing initial analysis
-
Preformed feature engineering by creating using bi-grams from byte files, pixel values from .asm files and selected the best features to optimize model performance
-
Conducted hyper-parameter tuning to increase performance of XGBoost model
Document Classification using CNN
-
Classified a total of 18828 text files into 20 different classes using 1D CNN with Tensorflow and Keras.
-
Preprocessed the text data using different python packages like regex to remove tabs, spaces, emails, numbers, special character and SpaCy for chunking the text
-
Embedding layer using Keras and pre-trained GloVe vectors were used to vectorize the text data and the results were compared.
New York City taxi trip duration prediction
-
Predicted the trip duration of taxi between locations in NYC using XGBoost Regressor.
-
Using the best model, a web app was created using streamlit.
-
Approach: Performed feature engineering to reduce the prediction error. Additional features were generated by adding extra data using table and spatial joins.
-
Use Case: Can be used in logistics domain for estimating the trip duration between two locations.