Dr Rishideep Roy

Research interests

Bayesian Statistics

The idea behind Bayesian statistics is making use of previously acquired knowledge as well as recent data to make better predictions. Sometimes we might take into account only the recent data, it might not capture previous or broader insights that one might have on the underlying phenomenons or models. The recent data might just be off. Or might happen that the basic nature of the underlying phenomenon has changed, and so, making a prediction based purely on previous insights is not effective. But Bayesian models bring previous knowledge into play and also take into account recent data to give an effective combined prediction. We have used these techniques in various applications like market shares of brands going through market churn, prediction of election results, and prediction of outcome of games.

Sports Analytics

I have been interested in sports analytics in general, and have worked on various sports like football, cricket, basketball, and tennis. I have worked on various projects, like predicting the outcome of a football match based on the events on the field, like goals, fouls, cards, corners, challenges, etc. We have also been working on refining this based on live commentaries and sounds captured in the live recording of the games. We have also been working on effective strategies for betting companies as well as their customers to find an optimal time to enter the bets, as well as decisions on whether to enter them or not. We have also been working on within-game substitution strategies in basketball to optimize the skills and form of players. We have further used sports analytics techniques to find the best starting line up in cricket, given a fixed pool of cricketers, as well as strategies for business franchises in auctions for an optimised team selection, to buy and sell players. I am generally interested in similar problems in other sports, as well as on other aspects of sports analytics.

Stochastic Processes

Log-correlated Gaussian fields are Gaussian processes with wide applications, from models in physics to financial data. Log-correlated Gaussian fields have been observed in a wide range of mathematical models, for example, the Random energy model, the Gaussian multiplicative chaos, as well the Circular Unitary Ensemble (CUE) of random matrix theory. They also appear in the maximum of Riemann Zeta function in number theory. They also have strong relations with Louville Quantum Gravity.

Interacting Particle Systems

I have worked on various stochastic models in the class of interacting particle systems. More popularly, some of them have been used to study spread of diseases, with various natures of diseases being studied by different models. One assumption about them is that these diseases spread through contact. In some, the diseased individual can spread it indefinitely, in some others they can recover and don't spread it anymore. In some models not everyone coming in contact with a disease will get infected. I have worked on various forms of these models. A very similar model in this regard is a model on spread of rumours, and how quickly they can flow.

Election Analytics

Elections in modern days are one of the most effective ways for citizens to voice their opinions. With democracy being the most common vehicle of governance, the utility of elections is ever-increasing. So, the process of electioneering has also generated more interest. We have used analytics in the prediction of elections, based on partial voting information. We have also developed techniques which can be used to detect electoral fraud. Similar techniques can be used to assess the effect that campaigns might have on the final electoral outcome, and whether they could tilt the result from one party's favor to the other or not.

Information Spread

I have worked on models related to spread of rumours and identifying keywords in communication spread in networks. Information in this current age in akin to oil, and hence, identification of it becomes extremely important in a variety of different fields like social media, government policy decision etc.

Time Series Analysis

I have worked on different aspects of time series, including forecasting in categorical and discrete time series. I have also worked on time series regression models, including non-parametric time series regression. We have extended current results in this field to include general spatio-temporal models.

Conferences and presentations

Within game forecasting in T20 cricket, a time series classification approach using parallelism

Invited presentation, EcoSta 2024, 7th International Conference on Econometrics and Statistics, Beijing, China, 18/7/2024

Co-existence in probability models

Invited presentation, 2022 IISA Conference, Bengaluru, India, 29/12/2022

Effect of media attention on crude oil price volatility using a non-parametric time series regression.

Invited presentation, 15th International Conference of the ERCIM WG on Computational and Methodological Statistics (CMStatistics 2022), London, United Kingdom, 18/12/2022

Effect of media attention on crude oil price volatility using a non-parametric time series regression.

Invited presentation, IMS International Conference on Statistics and Data Science (ICSDS), Florence, Italy, 15/12/2022

Coexistence in discrete time multi-type competing frog models

IMS Annual Meeting, June 2022, London, United Kingdom, 28/6/2022