A collection of assessments in Time Series Analysis completed as part of my Econometrics program.

Hybrid fuzzy model for financial time series forecasting

Data Science Capstone Project:

Non-linear topology identification using Deep Learning. Sparsity (lasso) is enforced in the sensor connections. The non-convex and non-differentiable function is solved using sub-gradient descent algorithm.

Stochastic Processes Comparison of Accuracy: Data Assimilation, Echo State Machines and Nonlinear Vector Autoregressive Learning Methods. ### Comparing discrete and variational data assimilation methods to reservoir computing - machine learning for synthetic chaotic nonlinear dynamical systems

This study is based on confirmed cases and deaths collected from Pakistan. Results demonstrate the promising potential of TIME SERIES model in forecasting COVID-19 cases and highlight the superior performance of the time series compared to the LSTM.we apply AI-based forecasting models such time series ARIMA, LSTM, prophet and VAR.

아주대학교 2021-2 비즈니스 애널리틱스 프로젝트

Filters (kalman, hodrick-prescott, moving average) together with comparison and sensitivity analysis (in notebook filters_with_parameters)+var analysis and granger causality test. Test for random walk (CE currencies using yfinance API)

Vector Autoregression augmented with deep learning.

The VAR model is used to forecast the appliances energy on the previous usage history. The data were first tested using adfuller test, granger casuality test. The lag value of 7 was determined for VAR model after running it iteratively for values upto 48.

*Time Series Forcasting Project* :- We tested Fb prophet model , VAR (Vector Autoregression) Model , Transformer Models(Only Encoder) and LSTM's.

MCMC estimation of Bayesian Vectorautoregressions

{mvgam} R 📦 to fit Dynamic Bayesian Generalized Additive Models for time series analysis and forecasting