Dense neural network with Optuna and XAI for the characterization of LdE-e adoption profiles

This code focuses on the classification of potential adoption profiles for the "Libro del Edificio Electrónico" (LdE-e). The project utilizes a Dense Neural Network (DNN) architecture optimized through Optuna for hyperparameter tuning. To address data imbalance, custom class weighting and probability threshold optimization were implemented. The study incorporates Explainable AI (XAI) techniques, including Permutation Feature Importance (PFI), Partial Dependence Plots (PDP), and SHAP (SHapley Additive exPlanations) to characterize the influence of psychological and socio-economic factors on LdE-e adoption. Additionally, the model is benchmarked against an optimized XGBoost classifier and subjected to stratified analysis by climate zones and energy expenditure levels.