Klasifikasi Data Kampanye Digital Marketing Menggunakan Random Forest dan Algoritma PSO

Ardiansyah, Mochamad Fachri (2025) Klasifikasi Data Kampanye Digital Marketing Menggunakan Random Forest dan Algoritma PSO. Undergraduate thesis, UPN Veteran Jawa Timur.

	Text (Cover) 20081010214.-cover.pdf Download (548kB)
	Text (Bab 1) 20081010214.-bab1.pdf Download (198kB)
	Text (Bab 2) 20081010214.-bab2.pdf Restricted to Repository staff only until 16 June 2027. Download (344kB)
	Text (Bab 3) 20081010214.-bab3.pdf Restricted to Repository staff only until 16 June 2027. Download (432kB)
	Text (Bab 4) 20081010214.-bab4.pdf Restricted to Repository staff only until 16 June 2027. Download (1MB)
	Text (Bab 5) 20081010214.-bab5.pdf Download (150kB)
	Text (Daftar Pustaka) 20081010214.-daftarpustaka.pdf Download (124kB)

Abstract

Digital marketing has become a key strategy for companies in facing an increasingly competitive marketing environment, yet the complexity of digital marketing campaign data makes analyzing its success a distinct challenge. Digital marketing faces complex challenges in analyzing campaign success in a competitive marketing environment. Campaign data classification requires an accurate approach to understand success factors. This research optimizes the Random Forest model using Particle Swarm Optimization (PSO) to improve the classification accuracy of digital marketing campaign data. The research method implements the Particle Swarm Optimization (PSO) algorithm to identify the most effective Random Forest parameter configuration, including the quantity of decision trees, maximum depth level, minimum threshold of samples for internal splits, and minimum sample constraints at leaf nodes. Model performance testing is measured with various evaluation indicators including accuracy, precision, recall, and ROC-AUC curves. The analysis results show that the Random Forest model optimized with PSO technique demonstrates significant improvement in its classification capabilities. The model achieves the highest accuracy at a 50% ratio with a value of 89.70%, precision of 89.72%, and ROC-AUC of 81.19%, with parameters of n-trees 14 decision trees, max_depth 22, min samples split 7, and min samples leaf 1. The main achievement is the reduction of computation time from 21.51 seconds to 0.48 seconds, with an accuracy improvement of 9.7% compared to the standard random forest model.

Item Type:

Thesis (Undergraduate)

Contributors: