Machine Learning Projects

Credit Fraud Detection | Python, Risk Analysis, Machine Learning

• Implemented a credit card fraud detection system using Logistic Regression, XGBoost, Decision Tree, SMOTE for class imbalance handling, achieving a ROC-AUC score of 0.97. Preprocessed data with feature scaling, PowerTransformer for skewed features, and evaluated model performance using precision-recall and ROC curves to optimize fraud detection strategies.

Credit Risk Modeling | Python, Risk Modeling, Machine Learning

• Implemented a credit risk model using machine learning algorithms - (Logistic Regression, Random Forest, XGBoost, SVC) to classify clients as default or non-default using data on credit repayment history and customer demographics, with XGBoost giving a better ROC score.

Engineering Cloud Computing: Skin Cancer Using AWS | Python , AWS

• Leveraged AWS cloud services to develop a robust and scalable skin cancer detection system using Python, incorporating Amazon Rekognition and Amazon SageMaker for enhanced diagnostic accuracy. Utilized AWS services such as S3 for data storage, IAM for secure access management, and Lambda for serverless processing to effectively address challenges related to scalability, cost, and global accessibility.

PCOS Classification | Python

• Developed a machine learning model to predict PCOS using a Kaggle dataset. Key features identified as significant predictors included follicle count, weight gain, cycle regularity, skin darkening, and hair growth. Achieved 91% accuracy using a Random Forest classifier.

Social Media Mining: Sentiment Analysis of Customer Reviews | Python , NLP

• Conducted sentiment analysis on over 80k Amazon mobile reviews using a Bi-LSTM deep learning model, achieving 88% accuracy. Applied data preprocessing techniques, including spelling correction and emoji handling, while effectively addressing class imbalance.

Smart Glasses for Indoor Object Detection to Assist Visually Impaired Using Raspberry Pi | Python , Raspberry Pi

• Developed a real-time object detection prototype using a camera connected to Raspberry Pi, providing audio feedback for visually impaired users. Leveraged TensorFlow API and COCO SSD MobileNet v1.

Classroom Scene Recognition for Monitoring Using MobileNet | Python

• Developed an indoor scene recognition system to classify environments such as concert halls, meeting rooms, and auditoriums. Initially employed GIST, DAISY, and HOG for feature extraction and SVM for classification, later enhancing accuracy with transfer learning using MobileNet and Keras.