Browsing by Author "Watson, Neil"

Now showing 1 - 3 of 3
  • No Thumbnail Available
    Item
    Open Access
    An affordable data solution for player recruitment for clubs in the South African Premier Soccer League
    (2025) King, Wesley; Watson, Neil
    As football becomes increasingly data-driven, the high cost of advanced player analytics threatens to leave resource-limited clubs at a competitive disadvantage, particularly in player scouting. This growing reliance on expensive, granular data under-scores the need for affordable, innovative data solutions. This dissertation seeks to democratize access to player evaluation data for football clubs in the South African Premier Soccer League. This is achieved by developing a cost-effective system that uses models to approximate Statsbomb's proprietary ‘On the ball' player evaluation metric using cheaper, frequency data from Wyscout and FBref. The analysis shows that linear regression models can effectively estimate key components of this metric using basic frequency statistics. The findings are then packaged into a prototype web-based Decision Support System with budget-aware scouting features, showcasing how club scouts and analysts can integrate sophisticated data-driven recruitment strategies into their clubs without incurring prohibitive data costs.
  • No Thumbnail Available
    Item
    Open Access
    Esports betting technology: machine learning for match prediction and odds estimation
    (2024) du Plessis, Henri Izak David; Watson, Neil
    Esports betting is a rapidly growing segment of the online sports betting market. A key feature of this industry is the pricing of betting odds. This study investigated the online sports betting industry, odds compilation, and how machine learning can be used for sports prediction. The techniques used in the literature were then applied to one of the world's foremost esports: Counter-Strike. A substantial dataset of professional match data (n=11271) was collected and used to construct 142 relevant features for match prediction. Several supervised learning models, including random forests, feed-forward neural networks, and XGBoost, were trained to estimate win probabilities for both teams in each match. Betting odds were then calculated using these probabilities and compared to real-world betting odds. A notable aspect of the research is the implementation of Microsoft's TrueSkill rating system. It served as both a benchmark and an input feature. Among the models tested, XGBoost showed the best overall performance. The highest match prediction accuracy attained was 62.7%. It was found that incorporating a large number of statistics did not significantly improve predictive accuracy when compared to models using fewer, more important features. It was also found that LAN matches and best-of-3 map formats are more predictable than their counterparts. Despite the inherent difficulty in Counter-Strike match prediction, the models could generate efficient odds which exhibited strong correlation with real-world odds (up to 85%). A betting strategy informed by the generated odds was back-tested over a six-month period and shown to be profitable. This research therefore demonstrates how machine learning models can be used for esports match prediction, with practical applications in the online betting industry.
  • Thumbnail Image
    Item
    Open Access
    Radar-Based Multi-Target Classification Using Deep Learning
    (2022) Mashanda, Nyasha Ernest; Watson, Neil; Gaffar, Yunus Abdul; Berndt, Robert
    Real-time, radar-based human activity and target recognition has several applications in various fields. Examples include hand gesture recognition, border and home surveillance, pedestrian recognition for automotive safety and fall detection for assisted living. This dissertation sought to improve the speed and accuracy of a previously developed model classifying human activity and targets using radar data for outdoor surveillance purposes. An improvement in accuracy and speed of classification helps surveillance systems to provide reliable results on time. For example, the results can be used to intercept trespassers, poachers or smugglers. To achieve these objectives, radar data was collected using a C-band pulse-Doppler radar and converted to spectrograms using the Short-time Fourier transform (STFT) algorithm. Spectrograms of the following classes were utilised in classification: one human walking, two humans walking, one human running, moving vehicles, a swinging sphere and clutter/noise. A seven-layer residual network was proposed, which utilised batch normalisation (BN), global average pooling (GAP), and residual connections to achieve a classification accuracy of 92.90% and 87.72% on the validation and test data, respectively. Compared to the previously proposed model, this represented a 10% improvement in accuracy on the validation data and a 3% improvement on the test data. Applying model quantisation provided up to 3.8 times speedup in inference, with a less than 0.4% accuracy drop on both the validation and test data. The quantised model could support a range of up to 89.91 kilometres in real-time, allowing it to be used in radars that operate within this range.