• Investigating Fouling Efficiency in Football Using Expected Booking (xB) Model
  • Proposed a novel metric “Expected Booking” (xB) to estimate the likelihood of a foul resulting in a card. The xB model can also be used to estimate fouling efficiency by comparing the expected bookings to the actual bookings received.
  • Model was trained and tested using StatsBomb event and 360 data for spatial context, employing ensemble methods to improve the model performance. The xB model incorporated ball possession value (VAEP) as a feature to quantify the threat posed by the team in possession.
  • Demonstrated the effectiveness of the xB model by using it to analyse the FIFA World Cup 2022 data. The model quantifies fouling efficiency of teams and players, identfiying the fouling efficiency and tactics of teams and players.
    
    
• Semantic-Guided Generative Image Augmentation with Diffusion Models: Implementation
  • Implemented and evaluated the Semantic-Guided Generative Image Augmentation Method with Diffusion Models (SGID) method to generate diverse and semantically consistent augmented images for image classification, using image labels, captions, and diffusion models, as proposed by Li et al. 2023.
  • Compared the proposed method with baseline augmentation techniques such as CutMix and RandAugment on various datasets and backbone architectures, and analyzed the results and challenges. SGID is capable of generating augmented images with more variability and semantic relevance, which helps the model generalize better to unseen data.
  • Replicated the original paper’s implementation details and experimented with different parameters and strategies to improve the quality and consistency of the generated images.
    
    
• PassValue: Valuing Passes using Football Tracking Data
  • Utilized tracking data to identify techniques for valuing passes, resulting in more comprehensive analysis than traditional events data as tracking data captures much more insights as it includes every player position at each instant.
  • Implemented a metric using tracking and event data to measure pass worth through calculations of pass distance, opponents positioning and press, pitch control, Valuing Actions by Estimating Probabilities (VAEP), Expected Possession Value (EPV) and pass errors. This metric can be used for ranking, scouting, and identifying teams and players with identical style.
    
    
• Conserve‑a‑BIT: Crowdfunding platform on Ethereum
  • Developed a decentralized crowdfunding platform on Blockchain that enables non-profits, government organizations, and startups to raise donations in a transparent and secure manner.
  • Designed and implemented an innovative smart contract for the platform that ensures transparency, safety, and eliminates third-party charges. The smart contract also empowers contributors with the authority to release funds, track fraudulent scammers, and reinvest their donations.
  • Utilized a diverse tech stack including Solidity for smart contract development, Node.js and React for backend and frontend development respectively, MongoDB for database management, and Web3.js for Ethereum JavaScript API.
    
    
• PlayFair: Online Trivia Games Helper
  • Developed a Python script that uses Optical Character Recognition (OCR) to read questions and options from online trivia games in real-time, automates web searches.
  • The script performs text analysis on search results to suggest the best answer in seconds, providing real-time assistance by suggesting the most likely correct answers.