• Conducted data exploration and preprocessing, managing multi-swipe, reversed duplicate transactions, missing values, feature encoding, normalization, feature selection, and data imbalance using bootstrapped iterative undersampling.
• Implemented hyperparameter tuning with RandomizedSearchCV across 4 ML models, achieving a 0.91 F-Beta Score with Gradient Boosting. Utilized DVC for ML pipeline staging and deployed the solution using FastAPI with Docker on AWS for efficient API.
• Leveraged the VGG16, and RESNET50 Model Transfer Learning, to extract key facial attributes from images. Performed Data Augmentation includes resize, zoom, contrast in training set. Images were then sent to VGG Model, converted into a 2D array with 2048 attributes in a vector.
• Devised a user-friendly Streamlit web portal matched preprocessed images with best match from list of feature vector on applying cosine similarity. Also employed Keras tuner to identify best optimizer (AdamW) and other parameters to accelerate efficient image matching.
In this Transfer Learning is performed by using models ResNet50, EfficientNetB0, and VGG16 for Video Classification and Frames Image Classification
This Biomedical data set was built by Dr. Henrique da Mota during a medical residenceperiod in Lyon, France. Each patient in the data set is represented in the data setby six biomechanical attributes derived from the shape and orientation of the pelvis and lumbar spine.
