Predicting Hand Strength and Bets in Poker with Machine Learning Ian Seymour, Alexander Anderson Project Mentor(s): Razvan Andonie, PhD Determining an optimal strategy for poker is difficult due to imperfect information. The strength of each player’s hand is hidden from others, leading to a disconnect between betting actions and true hand strength. Multiple machine learning methods were applied to Poker IRC data from Texas Hold’em games to predict player hand strength and betting behavior. K-Means was used to initially cluster the data, resulting in two clusters divided by players that bluff frequently and players that do not. Support Vector Regression (SVR), Random Forest, Multi-Layer Perceptron (MLP), and K-Nearest Neighbors machine learning models were compared via cross validation. SVR was found to be most effective. SVR was used to predict total bet amounts and hand strength. The results demonstrate that betting volume is highly predictable due to the relationship between player actions and betting. However, predicting hand strength was more challenging due to bluffing decoupling hand strength from betting actions. To improve results for predicting hand strength, we explored creating synthetic training data for the models. Presentation Type: Oral Presentation (May 20, 9:30am–5:00pm) Keywords : Machine Learning, Poker, Texas Hold’em, Player Profiling, K-Means SOURCE Form ID: 76 Machine Learning-Based Intrusion Detection: Logistic Regression and Naive Bayes for Probabilistic Analysis Ameli Tadzhibaeva Project Mentor(s): Chris Black, PhD This project designs a threat detection system using machine learning trained to analyze network traffic and identify malicious activity. In traditional cybersecurity environments, maintaining a balance between effective intrusion detection and minimizing false alarms is a challenge. While flagging malicious activity is crucial, false alarms can be misleading and overwhelming. The proposed system processes network traffic data and creates probabilistic predictions of potential intrusions, integrating two machine learning approaches: Logistic Regression and Naive Bayes Classifier. Logistic regression models the probability of an attack using a statistical learning approach based on a logistic function, while Naive Bayes estimates attack likelihood using Conditional Probability under the assumption of feature independence. The system is tested on a labeled dataset of network activity. This project examines how each model estimates attack likelihood. By evaluating both methods within a single detection framework, this study highlights the applicability of machine learning techniques to the development of an operational security monitoring system. Presentation Type: Oral Presentation (May 20, 9:30am–5:00pm) Keywords: Intrusion Detection, Machine Learning, Cybersecurity, Logistic Regression. Naive Bayes Classifier SOURCE Form ID: 224
114
Made with FlippingBook interactive PDF creator