Time Series Forecasting

Automated Cross-Validation Framework for Time Series Model Selection

Created an automated framework for a demand forecasting problem to choose the best-performing model on a weekly basis. The models considered are: Facebook Prophet Moving Average Simple Exponential Smoothing Holt Winter's ARIMA

Computer Vision

Medical Segmentation

Developed a model to automate the detection of Tumour in a MRI. Implemented the encoder-decoder-based deep learning architecture:- U-NET with Adam optimizer. Optimized the model by converting it to a TFLite Model to reduce the latency.

Image Classification

Plant Seedling Classification (Kaggle Competition)

Used pre-trained VGG16, ResNet50, and InceptionV3 networks to extract bottleneck features and build a model on top of each of them to evaluate and compare the model performances. (Model performances include classification report, confusion matrices, plots of Loss Vs Epochs) Optimized models using TF-Lite (Post Training Dynamic range quantization)

Classification

Credit Card Default and LIME

Used tree-based algorithms (Decision tree, Random Forest, XGBoost) to determine whether a client will default or not. Explored LIME library and concepts like PD plots, local interpretation, surrogate models to explain the overall as well as instance-wise model performance.