My Academic Projects

Credit Risk Management

Engineered credit card risk assessment strategies by employing advanced XGBoost and Neural Network ML models.
Processed an extensive dataset of over 50,000 records, emphasizing feature engineering and one-hot encoding to enhance model efficacy.
Defined conservative and aggressive portfolio strategies aim to strike a balance between risk tolerance and optimal portfolio performance while adhering to the stringent default rate constraint of 10%.

Hospital Readmission Prediction

Used R for data preprocessing, statistical analysis, and visualization on Kaggle dataset with 25,000+ records.
Executed logistic regression and decision tree models, achieving respective accuracies of 61.34% and 60.25%.

Avocado Weekly Sales - Forecasting

The dataset includes data on daily avocado sales by bag size and type (conventional or organic) across various US locations over six years. With over 30,000 entries, it covers essential columns like date, average price, total volume, and bag sizes, providing a thorough insight into avocado sales dynamics.
For forecasting future values, the Holt-Winters method, known for its capability in handling trends and seasonality, was selected.
Experimentation with different combinations of trend and seasonality parameters led to the identification of the optimal model for predicting sales over the next two weeks.

Convolution Neural Network for Classification of Cat and Dog

* Implemented Convolution Neural Network algorithm by adding three layers of convolution to extract the feature from the image and predict whether the image is of a Cat or a Dog.
* Trained and tested the model using a total of 2500 images of cats and dogs, which attained a prediction accuracy of 87%

Hosting a website on Amazon Web Services

▪ Used HTML and CSS to design a user-friendly website named Adventure Quest for storing crucial data on a local phpMyAdmin server.
▪ Launched an instance using the Elastic Compute Cloud and stored it using the Simple Storage Service.
▪ Generated and managed more than 50 records, which were developed and put into production on an AWS instance

Adaptive Technology using Morse code in Hindi language

▪ Generated speech and letters in Devanagari using morse code for native language speakers with different limitations.
▪ Created a decryption function to translate Morse to Hindi script and built a python dictionary for 48 Hindi letters equivalent to morse code.
▪ Obtained proper Hindi script using morse code with 75% accuracy, with a user-friendly GUI.

Text Classification on Movie review

▪ Collectively worked in a group of two to analyze more than 200 Twitter reviews using Naïve Bayes’ classifier for the probabilistic approach to review the movie holistically.
▪ Pre-Processed the data using NLTK to remove slang words, tokenization, and normalize the case, and created a feature vector that extracts specific features from the tweets.
▪ The Naïve Bayes classifier was trained with over 1600 positive and negative reviews, and the results were visualized using a Pie chart.