MSc Proposal " Advance Multivariate Multi-Time-Series Framework for a Novel COVID-19 Dataset" By Swastik Bagga

The School of Computer Science is pleased to present…

Thesis Proposal Announcement

MSc Thesis Proposal by: Swastik Bagga

Date: Wednesday, August 30, 2023

Time: 2:00 pm-4:00 pm

Location: Essex Hall Room 122

This thesis introduces a novel framework tailored for the intricate challenges posed by multivariate multi time

series datasets in regression prediction. Demonstrated through a novel COVID-19 dataset, the framework not only

advances the field's approach to complex predictive analytics when underling timeseries are very complex but also

offers accurate predictions of epidemic trends at regional or provincial levels, overcoming limitations of nationallevel

analysis.

The proposed framework integrates advanced data preprocessing, feature selection, feature engineering, feature

encoding and model architecture to address intricate variable interactions and temporal dependencies. Its

application to COVID-19 forecasting showcases the framework's ability to provide targeted and reliable predictions

crucial for decision-making. Solving multivariate multi time series regression problems is daunting due to

intertwined variables and temporal dynamics. The presented framework tackles this complexity by effectively

capturing dependencies and latent patterns while considering real-world uncertainties, thus opening avenues for

robust predictions across various domains. By harnessing the multivariate nature of the data, the framework

improves prediction accuracy and empowers decision-makers with targeted insights.

The proposed framework rises to this challenge by offering a cohesive solution that captures latent patterns,

models dependencies, and embraces the uncertainties present in real-world data. This work not only extends the

boundaries of predictive analytics but also underscores its practical significance in informing policies and strategies,

particularly in the context of localized epidemic trend forecasts.

After initial experimentation proposed framework has been observed to beat tradition regression models such as

KNN, Random Forest, and traditional time series models such as ARIMA/VAR models on metrics such as R^2, MAE

(Mean Absolute Error) and RMSE (Root Mean Square Error).

Keywords: Deep Learning, Multi Variate Time Series, Multi Variate Multi Timeseries Dataset, COVID-19

Epidemic Trends, Deep Learning Encoding.

 

Internal Reader: Dr. Saeed Samet

External Reader: Dr. Ahmed Hamdi Sakr

Advisor: Dr. Ziad Kobti