KickClone: A Machine Learning Model Built to Revolutionize Football Scouting

This research paper presents KickClone, a football player replacement model that employs cosine similarity as the primary method, along with a machine learning approach in the form of Principal Component Analysis (PCA) for dimensionality reduction, for identifying optimal substitutes. Utilizing a huge dataset of over 200,000 players from EAFC 24, KickClone compares player profiles based on various traits and characteristics, providing an advanced solution for scouting and player replacement. Unlike traditional scouting methods or basic statistical matching tools, KickClone uniquely combines cosine similarity with PCA-based dimensionality reduction to deliver a high-resolution, attribute-driven comparison across a vast player space, enabling nuanced and precise player replacements. We have performed thorough data preprocessing to normalize key numerical features like a player’s pace, shooting, and passing, ensuring comparability. Categorical features, including player positions, have been encoded numerically. KickClone calculates the cosine similarity between player vector embeddings, focusing greatly on attribute-based proximity. This approach eliminates external influences, precisely matching a player’s skill and overall performance. The model identifies players with high cosine similarity to any specified player in the dataset and then gives the optimal replacements after calculating the Euclidean distance as a second metric. The results demonstrate the model’s effectiveness in enhancing the painstaking process of finding player replacements. The model can offer football clubs and scouts a computationally efficient and accurate tool for scouting. KickClone represents a contemporary and data-centric approach for better squad management.