A Real-Time Analysis of Predicting Titanic Survival Using Machine Learning Models

Abstract

Survival prediction was a popular data mining task, providing insights in data-driven decision-making. The present study proposed a systematic approach for predicting survival using machine learning models, explainable artificial intelligence methods, and real-time inference. A robust preprocessing framework was developed to handle missing data, encode categorical variables, and address class imbalance, enhancing data integrity and quality. Data engineering methods, such as the creation of family size and encoding of age groups, were used to extract relevant insights from the data. Various classification algorithms, including Logistic Regression, Decision Tree, Random Forest, Support Vector Machine, and Gradient Boosting, were applied with cross-validation for optimal tuning. Models were evaluated based on accuracy, precision, recall, F1-score, and ROC-AUC, allowing for a comprehensive evaluation of their performance. Ensemble models were found to be most effective, as they model complex interactions between features. For improved transparency and understanding, explainable artificial intelligence techniques such as SHAP and LIME were employed to understand the contribution of features to predictions. We also developed a real-time predictive system by combining preprocessing with the trained model to enable fast prediction of new data. The framework was tested for accuracy and speed, showing its potential for real-world use. The results underline the need for a balance of accuracy, interpretability, and real-time prediction to build robust and scalable models.