Reduksi Dimensi Metode Principal Component Analysis dan Linear Discriminant Analysis Data Gambar dengan Scilab: Dimensionality Reduction of Image Data Using Principal Component Analysis and Linear Discriminant Analysis in Scilab

Ariadi Retno Tri Hayati Ririd (Politeknik Negeri Malang)

Jurnal Pendidikan Sains dan Komputer
Vol. 6 No. 03 (2026): Call for Papers Oktober 2026

This research involves two learning stages: the first uses Principal Component Analysis (PCA) to reduce training dimensions, followed by a second stage using Linear Discriminant Analysis (LDA). Experiments were conducted on the Olivetti dataset (in grayscale format) using the Scilab programming language, with the data partitioned into training and testing sets. The results show a significant increase in classification accuracy despite using very few dimensions. In the first experiment (using 32 data points), accuracy improved by 18.75%—rising from 68.75% (using PCA with 8 dimensions) to 87.5% (using LDA with 7 dimensions). In the second experiment (using 100 data points), accuracy increased by 16%—from 84% (PCA) to 100% (LDA). This process reduced the original feature space—which spanned thousands of dimensions—to 40 dimensions via PCA, and then to 9 dimensions via LDA, using 50 training samples and 50 testing samples. In the third experiment (using 240 data points), PCA achieved 67.5% accuracy, while LDA reached 85% using 23 dimensions. Overall, the average accuracy across the test sets was 73.41% for Principal Component Analysis and 90.83% for Linear Discriminant Analysis.