Implementasi Machine Learning untuk Memprediksi Energi Ikat Inti Atom dengan Algoritma Generalized Additive Model dan Explainable Neural Network Berbasis Generalized Additive Model

Kristiyan Laoli, Drs. Pekik Nurwantoro, M.S., Ph.D.

2023 | Skripsi | S1 FISIKA

Telah diprediksi energi ikat inti atom dengan menggunakan dua model machine learning yaitu Generalized Additive Model (GAM) dan Explainable Neural Network Berbasis Generalized Additive Model (GAMI-Net). Fitur-fitur yang digunakan dalam kedua model ini meliputi properti inti atom seperti proton (Z), neutron (N), nomor massa atom (A), jejari, permukaan, dan |N-Z|, serta juga meliputi notasi inti seperti magicZ (posisi proton di dalam magic number), magicN (posisi neutron di dalam magic number), Z_valence (proton pada kulit terluar), N_valence (neutron pada kulit terluar), dan pair (bilangan ganjil atau genap dari Z dan N). Hasil dari GAM menunjukkan RMSE sebesar 0,300 MeV dengan waktu latih 14,47 detik, sedangkan GAMI-Net menunjukkan RMSE sebesar 0,481 MeV dengan waktu latih 1959.65 detik (32,6 menit). Selain itu, interpretasi fitur, yaitu interpretasi setiap properti inti dan notasi inti terhadap energi ikat inti berdasarkan kedua model, telah berhasil diperoleh. Pendekatan machine learning dapat menjadi alternatif untuk menjelaskan energi ikat inti atom dengan mempertimbangkan properti inti dan notasi inti, yang dapat membantu dalam pemahaman lebih lanjut tentang sifat inti atom.

The binding energy of atomic nuclei has been predicted using two machine learning models, namely the Generalized Additive Model (GAM) and the Explainable Neural Network based on Generalized Additive Model (GAMI-Net). The features utilized in both models include atomic nucleus properties such as proton (Z), neutron (N), atomic mass number (A), radius, surface area, and |N-Z|, as well as nucleus notation such as magicZ (proton position within the magic number), magicN (neutron position within the magic number), Z_valence (proton in the outermost shell), N_valence (neutron in the outermost shell), and pair (even or odd number of Z and N). The GAM results showed an RMSE of 0.300 MeV with a training time of 14.47 seconds, while GAMI-Net exhibited an RMSE of 0.481 MeV with a training time of 1959.65 seconds (32.6 minutes). Furthermore, feature interpretation, which refers to the interpretation of each atomic property and nucleus notation in relation to the binding energy of the nucleus based on both models, was successfully achieved. The machine learning approach can serve as an alternative for explaining the binding energy of atomic nuclei by considering nucleus properties and notation, thus aiding in further understanding of atomic nucleus characteristics.

Kata Kunci : energi ikat inti, machine learning, generalized additive model, explainable neural network, interpretasi fitur.