NAÏVE BAYES OPTIMIZATION FOR VISUAL ERGONOMICS PREDICTION IN SMARTPHONE DISPLAY MODE

TEGAR PAMBUDI, RIZKHA (2026) NAÏVE BAYES OPTIMIZATION FOR VISUAL ERGONOMICS PREDICTION IN SMARTPHONE DISPLAY MODE. S1 thesis, Universitas Ngudi Waluyo.

	Text Cesillia Ayu Kumala Sari_144221003_artikelpdf - Cesillia Ayu Kumala Sari.pdf Restricted to Registered users only Download (916kB)
	Text Cesillia Ayu Kumala Sari_144221003_pengesahanartikel - Cesillia Ayu Kumala Sari.pdf Download (143kB)
	Text Cesillia Ayu Kumala Sari_144221003_lampirandepan - Cesillia Ayu Kumala Sari.pdf Restricted to Registered users only Download (998kB)
	Text Cesillia Ayu Kumala Sari_144221003_bab_I - Cesillia Ayu Kumala Sari.pdf Download (80kB)
	Text Cesillia Ayu Kumala Sari_144221003_bab_III - Cesillia Ayu Kumala Sari.pdf Download (86kB)
	Text Cesillia Ayu Kumala Sari_144221003_bab_V - Cesillia Ayu Kumala Sari.pdf Download (99kB)
	Text Cesillia Ayu Kumala Sari_144221003_daftarpustaka - Cesillia Ayu Kumala Sari.pdf Restricted to Registered users only Download (98kB)
	Text Cesillia Ayu Kumala Sari_144221003_fulltextpdf - Cesillia Ayu Kumala Sari.pdf Restricted to Registered users only Download (4MB)
	Text Cesillia Ayu Kumala Sari_144221003_fulltextdoc - Cesillia Ayu Kumala Sari.docx Restricted to Registered users only Download (3MB)
	Text Cesillia Ayu Kumala Sari_144221003_lembarkonsul - Cesillia Ayu Kumala Sari.pdf Download (431kB)

Abstract

Abstract−Smartphone display mode plays a significant role in preserving visual comfort and limiting the danger of digital eye strain due to prolonged use. This work attempts to maximize the effectiveness of the Naïve Bayes algorithm to predict user display mode preferences based on seven visual ergonomics characteristics, including age, gender, duration of usage, dominant time, purpose of use, screen type, and lighting circumstances. Data was acquired through an online poll of 283 respondents using convenience sample procedures. The main challenge faced was the imbalance in the initial data class distribution, with a ratio of 63% dark mode preference to 37% light mode, which could potentially cause prediction bias in the standard model. To tackle this, a dual optimization technique was applied: the Synthetic Minority Over-sampling Technique for Nominal (SMOTEN) to balance the data, along with GridSearchCV to discover the appropriate hyperparameters. The evaluation results indicated a considerable improvement in performance; the baseline model had an accuracy of 68.42% with a light mode recall value of 0.57, while the optimal model had an accuracy of 70.18% with a significant increase in recall to 0.71. This improvement indicates that the integration of SMOTEN significantly decreases majority prejudice, resulting in a more sensitive and trustworthy model as a foundation for the creation of adaptive interfaces directed towards visual health.

Actions (login required)

View Item