Identifikasi Big Five Personality pada Tangkapan Layar Wawancara dengan Metode Deep Ensemble

Fajri, Beauty Valen (2025) Identifikasi Big Five Personality pada Tangkapan Layar Wawancara dengan Metode Deep Ensemble. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5025211227-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (19MB) | Request a copy

Abstract

Perkembangan teknologi digital telah mendorong integrasi Artificial Intelligence (AI) dalam menganalisis data non-struktural, seperti citra, untuk berbagai tujuan, termasuk identifikasi kepribadian. Penelitian ini bertujuan untuk menganalisis hubungan antara representasi visual dalam gambar wawancara daring dengan karakteristik kepribadian berdasarkan model Big Five Personality, yaitu Extraversion, Agreeableness, Conscientiousness, Neuroticism, dan Openness. Ekstraksi fitur dilakukan menggunakan arsitektur ResNet50, yang menghasilkan 2048 dimensi fitur representatif. Fitur-fitur ini kemudian digunakan sebagai input dalam pemodelan prediksi kepribadian dengan pendekatan Deep Ensemble, yaitu Deep Super Learner, yang terdiri dari lima Base Learners (Random Forest, Extra Trees, Ridge, Support Vector Regression, dan XGBoost) dan satu Meta Learner (Hist Gradient Boosting Regressor). Penelitian ini melakukan dua eksperimen, eksperimen pertama menggunakan fitur penuh, yaitu 2048 dimensi, sedangkan eksperimen kedua menggunakan fitur yang telah direduksi menjadi 230 komponen utama. Evaluasi dilakukan dengan berbagai metrik, antara lain RMSE, MAE, R^2, Explained Variance, Max Error, Median AE, dan Standard Deviation. Hasil menunjukkan bahwa model Extra Trees pada eksperimen pertama dengan nilai RMSE 0,12400, MAE 0,09945, R^2 0,28616, Explained Variance 0,28715, Max Errror 0,39569, Median AE 0,08403, dan Standard Deviation 0,12391. Model Support Vector Regression pada eksperimen kedua dengan nilai RMSE 0,12371, MAE 0,09901, R^2 0,28803, Explained Variance 0,28989, Max Errror 0,40900, Median AE 0,08388, dan Standard Deviation 0,12355, memberikan performa terbaik di antara Base Learners, sedangkan Meta Learner Hist Gradient Boosting Regressor pada eksperimen pertama dengan nilai RMSE 0,12403, MAE 0,09942, R^2 0,28573, Explained Variance 0,28670, Max Errror 0,39614, Median AE 0,08412, dan Standard Deviation 0,12395 serta pada eksperimen kedua dengan nilai RMSE 0,12598, MAE 0,10070, R^2 0,26348, Explained Variance 0,26411, Max Errror 0,41344, Median AE 0,08621, dan Standard Deviation 0,12593. Perbandingan antar eksperimen menunjukkan bahwa reduksi fitur dapat meningkatkan performa base learners, namun sebaliknya menurunkan performa meta learner.
================================================================================================================================
The advancement of digital technology has driven the integration of Artificial Intelligence (AI) in analyzing unstructured data, such as images, for various purposes, including personality identification. This study aims to analyze the relationship between visual representations in online interview images and personality characteristics based on the Big Five Personality model, namely Extraversion, Agreeableness, Conscientiousness, Neuroticism, and Openness. Feature extraction was conducted using the ResNet50 architecture, which produced 2048 representative feature dimensions. These features were then used as input for personality prediction modeling using a Deep Ensemble approach, specifically Deep Super Learner, which consists of five Base Learners (Random Forest, Extra Trees, Ridge, Support Vector Regression, and XGBoost) and one Meta Learner (Hist Gradient Boosting Regressor). Two experiments were conducted: the first used the full 2048-dimensional features, while the second used reduced features comprising 230 principal components. The models were evaluated using various metrics, including RMSE, MAE, R^2, Explained Variance, Max Error, Median AE, and Standard Deviation. Results showed that the Extra Trees model in the first experiment achieved the best performance among base learners with RMSE of 0.12400, MAE of 0.09945, R^2 of 0.28616, Explained Variance of 0.28715, Max Error of 0.39569, Median AE of 0.08403, and Standard Deviation of 0.12391. The Support Vector Regression model in the second experiment also showed strong performance with RMSE of 0.12371, MAE of 0.09901, R^2of 0.28803, Explained Variance of 0.28989, Max Error of 0.40900, Median AE of 0.08388, and Standard Deviation of 0.12355. Meanwhile, the Meta Learner Hist Gradient Boosting Regressor in the first experiment recorded RMSE of 0.12403, MAE of 0.09942, R^2 of 0.28573, Explained Variance of 0.28670, Max Error of 0.39614, Median AE of 0.08412, and Standard Deviation of 0.12395; while in the second experiment, it obtained RMSE of 0.12598, MAE of 0.10070, R^2 of 0.26348, Explained Variance of 0.26411, Max Error of 0.41344, Median AE of 0.08621, and Standard Deviation of 0.12593. A comparison between the two experiments indicates that feature reduction can improve the performance of base learners, but conversely reduce the performance of the meta learner.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Prediksi kepribadian, Wawancara Daring, Analisis Gambar, Big Five Personality, Base learners, ResNet50, Deep Super Learner. Personality Prediction, Online Interview, Image Analysis, Big Five Personality, Base learners, ResNet50, Deep Super Learner.
Subjects:	Q Science > QA Mathematics > QA336 Artificial Intelligence
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Informatics Engineering > 55201-(S1) Undergraduate Thesis
Depositing User:	Beauty Valen Fajri
Date Deposited:	29 Jul 2025 02:24
Last Modified:	29 Jul 2025 02:24
URI:	http://repository.its.ac.id/id/eprint/122208

Actions (login required)

View Item