Age-Wise and Gender-Wise Emotion Prediction Classification Using The E-AADEP Framework

Abstract

This paper presents a comprehensive demographic analysis of the Enhanced Adaptive Attention-Based Dream Emotion Predictor (E-AADEP) framework, focusing on age-wise and gender-wise classification of dream emotions from EEG-derived features in the emotions.xls dataset (n = 2,131 samples, 2,548 features, 3 classes: NEGATIVE, NEUTRAL, POSITIVE). Participants were stratified into four age groups (A1: 0–16, A2: 17–30, A3: 31–45, A4: 46+) and two gender groups (Male, Female) using population-realistic proportions drawn from sleep-EEG demographic literature. A Random Forest ensemble (T = 200 trees, Gini impurity splitting) was trained on the full 2,548-dimensional feature vector and evaluated separately within each demographic subgroup. Overall test accuracy reached 99.30% (AUC-ROC = 0.9998). Age-wise accuracy ranged from 98.39% (A1) to 100.00% (A2), and gender-wise accuracy was 99.00% (Male) and 99.56% (Female). Chi-square tests confirmed that classification correctness is statistically independent of both age group (χ² = 1.45, p = 0.694) and gender (χ² = 0.012, p = 0.912), indicating no systematic demographic bias. Per-class F1-scores, confusion matrices, and AUC-ROC values are reported for every demographic subgroup, together with a combined age × gender accuracy heatmap.