Comparison of Recurrent Neural Network and Naive Bayes Algorithms in Identifying Stunting in Toddlers

Abstract

Stunting in toddlers is a health issue that affects their quality of life. This study aims to predict stunting status using three classification methods: Recurrent Neural Network (RNN), Long Short-Term Memory (LSTM), and Gaussian Naive Bayes. The dataset from Kaggle was split into 70% for training and 30% for testing to ensure optimal model evaluation. The RNN model was built with three hidden layers of 64 units each, while the LSTM model had four hidden layers with the same number of units. Both models utilized hidden states to capture temporal patterns and employed the tanh activation function to detect complex data patterns. The ADAM optimizer with a learning rate of 0.001 was applied to accelerate convergence. In contrast, the Gaussian Naive Bayes model used a simple probabilistic approach without temporal patterns, making it suitable for simpler datasets. Evaluation using accuracy and RMSE showed that LSTM achieved the highest accuracy (91%), followed by RNN (90%), though both exhibited signs of overfitting. Gaussian Naive Bayes attained 72% accuracy with stable performance. While LSTM and RNN effectively capture complex temporal patterns, they are prone to overfitting, whereas Gaussian Naive Bayes is suitable for initial implementation or simpler datasets, supporting early intervention for stunted toddlers.