Intrusi merupakan suatu tindakan yang dilakukan seseorang untuk menyusup, mencuri dan merusak sebuah sistem melalui jaringan internet. Beberapa jenis intrusi yang paling umum terjadi diantaranya adalah brute force, Denial of Service (DoS), Distributed Denial of Service (DDoS), infiltration, web attack, botnet, dan portscan. Salah satu mekanisme keamanan jaringan untuk mendeteksi intrusi yang ada saat ini adalah Intrusion Detection System (IDS). Namun karena perkembangan data yang kian pesat, mengakibatkan model IDS yang ada sekarang kurang efektif dalam mendeteksi intrusi. Oleh karena itu, para ahli melakukan penelitian untuk menggunakan pendekatan data mining dalam membangun model IDS. Pada penelitian ini, dilakukan perbandingan kinerja antara algoritma backprogation standar atau gradient-descent (GD), Levenberg-Marquardt (LM) dan Support Vector Machine (SVM) dalam membangun model IDS dengan dataset CICIDS2017 melalui supervised learning. Kinerja diukur berdasarkan nilai precision, accuracy, detection rate, dan False Alarm Rate (FAR). Dari penelitian diperoleh hasil bahwa algoritma GD memiliki kinerja paling buruk. Nilai FAR algoritma LM pada skenario brute force, Denial of Service (DoS), web attack, infiltration, botnet, Distributed Denial of Service (DDoS), dan portscan secara berurutan adalah 0.73%, 1.36%, 1.15%, 7.69%, 1.41%, 14.47%, dan 0.21%, sedangkan nilai detection rate-nya adalah 99.45%, 99.61%, 99.39%, 90.91%, 99.49%, 99.61% dan 99.88%. Nilai FAR algoritma SVM secara berurutan adalah 0.52%, 1.6%, 2.17%, 0%, 3.25%, 14.59%, dan 0.04%, sedangkan nilai detection rate-nya adalah 99.64%, 99.73%, 98.93%, 72.73%, 98.81%, 99.77%, dan 99.95%. Dari hasil penelitian, algoritma LM sedikit lebih unggul dibandingkan dengan algoritma SVM. Kata kunci: Intrusion Detection System, Levenberg-Marquardt, Support Vector Machine, Jaringan Saraf Tiruan, supervised learning Intrusion is an activity which was used by a person to intrude, steal and destruct a system through the internet network. Some common intrusion types that often occured are brute force, Denial of Service (DoS), Distributed Denial of Service (DDoS), infiltration, web attacks, botnets, and portscan. Nowadays, there is a networking security mechanism to detect intrusions, namely Intrusion Detection System (IDS). Because the data has increased rapidly, it impact to the conventional IDS model less effective in detecting intrusion. Therefore, some of academic scholars have conducted a research that uses a data mining approach in developing IDS models. In this study, a performance comparison between standard backprogation algorithms or gradient-descent (GD), Levenberg-Marquardt (LM) and Support Vector Machine (SVM) was conducted in developing IDS models using CICIDS2017 dataset through supervised learning. The performance was measured based on precision, accuracy, detection rate, and False Alarm Rate (FAR) values. The results emperically show that the GD algorithm has the worst performance. The FAR values of the LM algorithm for brute force, Denial of Service (DoS), web attacks, infiltration, botnet, Distributed Denial of Service (DDoS), and portscan were 0.73%, 1.36%, 1.15%, 7.69%, 1.41% , 14.47%, and 0.21% respectively, whereas the detection rate values were 99.45%, 99.61%, 99.39%, 90.91%, 99.49%, 99.61% and 99.88% respectively. The FAR values of the SVM algorithm were 0.52%, 1.6%, 2.17%, 0%, 3.25%, 14.59%, and 0.04% respectively, whereas the detection rate values were 99.64%, 99.73%, 98.93%, 72.73%, 98.81% , 99.77%, and 99.95%. According to the evaluation results, the LM algorithm was slightly superior compared to the SVM algorithm. Keywords: Intrusion Detection System, Levenberg-Marquardt, Support Vector Machine, Artificial Neural Networks, supervised learning