CONTINUAL LEARNING APPROACH TO EARLY PREDICTION STUNTING IN INFANTS AND TODDLERS AGED 0-5 YEARS

• Main Author: Faisal Moh Al Zein, Dicky
• Format: Theses
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/76199
• Institution: Institut Teknologi Bandung

Stunting is an event that inhibits the process of growth and development of children due to poor nutrition, repeated infections, and unfulfilled psychosocial stimulation. Stunting is still one of the biggest nutritional problems in the world, including Indonesia. Statistical analyses such as logistic regression, mixed regression, multivariate logistic regression and machine learning such as decision tree, naïve bayes, Support Vector Machine, extreme gradient boosting, Projective Adaptive Resonance Theory and Artificial Neural Network are widely used for stunting detection. In statistical analysis, simulated retrieval is still manual while machine learning is unable to adapt to new conditions and performance tends to decrease. In this study, a new approach to stunting detection is proposed using continual learning methods with the PackNet algorithm. Data from IDHS is processed in several stages including merge PR records with IR, outlier tests, multicollinearity tests, and logistic regression tests. Variables that significantly af ect stunting were selected to test the algorithm. The selection of variables uses logistic regression with a significant value of 5%, which is processed using SPSS. Based on research conducted using data from IDHS models trained with the PackNet algorithm can predict stunting ef iciently. The resulting performance when learning data flows from IDHS is getting better and is able to adapt when given some new tasks.