Predictive Modelling for Life Expectancy Classification: A Comprehensive Comparative Analysis of Ensemble Learning and Neural Network Approach

The research paper aims to showcase an in-depth investigation into predictive modelling for life expectancy classification, using advanced machine learning techniques. The research studies a comprehensive dataset consisting a wide range of socio-economic, health and demographic indicators from various countries. Including diverse factors such as vaccination coverage, disease prevalence, lifestyle choices and economic parameters, the dataset offers a rich pool of information for understanding the determinants of life expectancy. With features ranging from immunization rates and disease incidence to GDP per capita and education metrics, this dataset provides a holistic view of the multifaceted factors influencing public health outcomes. The study covers a broad spectrum of algorithms, including traditional models like Logistic Regression, K-Nearest Neighbours, Support Vector Machine, Naive Bayes, Decision Tree, as well as sophisticated ensemble techniques such as Random Forest, AdaBoost and XGBoost. The investigation further incorporates Neural Networks, specifically Multilayer Perceptron (MLP), leveraging the sequential model provided by Keras. With careful hyperparameter tuning and feature engineering, the models have been optimized to achieve the highest accuracy. Starting with a pre-processing phase, and then moving down to enclosing data down sampling, feature selection and standardization the research aims to provide the best classification model on the dataset. Subsequently, a suite of ensemble learning algorithms is applied, each fine-tuned through grid search and cross-validation. The investigation extends to the integration of neural networks, exploring different architectures and activation functions to enhance predictive accuracy. Outcome highlights the efficacy of AdaBoost as the best-performing ensemble model, and the MLP neural network gives competitive accuracy. The study also reveals insights into the impact of hyperparameter tuning on model performance. Furthermore, an exploration of the dataset structure prompts the incorporation of additional layers and activation functions in the MLP, achieving noteworthy improvements in accuracy. The study is concluded by presenting a comprehensive analysis of model performance, evaluating recall, precision, confusion matrices and Receiver Operating Characteristic (ROC) curves. The findings contribute valuable insights into the selection of optimal predictive models for life expectancy classification, paving the way for enhanced healthcare analytics and decision-making.