SUPRIYADI, Didi and Purwanto, Purwanto and Warsito, Budi (2024) MODEL KLASIFIKASI KINERJA AKADEMIK BERBASIS MULTIDIMENSIONAL ACADEMIC INFLUENCE FRAMEWORK DAN ARTIFICIAL NEURAL NETWORKS ALGORITHM. Doctoral thesis, UNIVERSITAS DIPONEGORO.

	Text 0. HALAMAN DEPAN_DISERTASI_DIDI SUPRIYADI.pdf Download (7MB)
	Text BAB I_DISERTASI_DIDI S.pdf Download (1MB)
	Text BAB II_DISERTASI_DIDI S.pdf Download (2MB)
	Text BAB III_DISERTASI_DIDI S.pdf Restricted to Repository staff only Download (1MB)
	Text BAB IV_DISERTASI_DIDI S.pdf Restricted to Repository staff only Download (2MB)
	Text BAB V_DISERTASI_DIDI S.pdf Restricted to Repository staff only Download (388kB)
	Text 2. DAFTAR PUSTAKA_Disertasi_DIDI SUPRIYADI.pdf Download (4MB)
	Text 3. LAMPIRAN_DISERTASI_DIDI SUPRIYADI.pdf Restricted to Repository staff only Download (5MB)

Abstract

Perguruan Tinggi mengatur predikat akademik mahasiswa melalui capaian kinerja akademik mahasiswa, salah satunya dengan waktu kelulusan. Kinerja akademik mahasiswa dipengaruhi oleh multi-faktor dari berbagai aspek atau dimensi yang menjadi pertimbangan. Dimensi yang mempengaruhi kinerja akademik mencakup dimensi personal mahasiswa, dimensi keluarga, dan dimensi kualitas layanan perguruan tinggi. Faktor – faktor dari ketiga dimensi tersebut diukur menggunakan kuesioner yang selanjutnya disebut dengan kerangka kerja Multidimensional Academic Influence Framework (MAIF). Pemantauan kinerja akademik penting untuk untuk mencegah kegagalan akademik yang dapat berdampak negatif pada penyelenggara Pendidikan maupun mahasiswa. Pengelolaan kinerja akademik membantu perguruan tinggi dalam pengambilan keputusan dan meningkatkan keunggulan kompetitif. Hal ini juga mempengaruhi kebijakan akademik perguruan tinggi, perbaikan kurikulum, hingga modifikasi pedagogik dosen.
Model Dimensions Student Intelligence (DSI) sebagai model klasifikasi biner untuk mengklasifikasi kinerja akademik mahasiswa dalam salah satu dari dua kelas yang berbeda menggunakan Multidimensional Academic Influence Framework (MAIF) dan algoritma Artificial Neural Network (ANN). Model ini dioptimasi pada prosedur pemodelannya melalui tahap pra-pemrosesan data dengan teknik pemilihan fitur menggunakan feature selection dan teknik handling imbalanced dataset
menggunakan resampling. Teknik Feature Selection dilakukan menggunakan metode Hibrida Spearman Rho dan Recursive Feature Elimination (RFE) untuk menentukan fitur yang berkorelasi terhadap waktu lulusan mahasiswa dan relevan
terhadap akurasi dan efisiensi model. Teknik handling imbalanced dataset dilakukan menggunakan metode Random Oversampling (ROS) yang menambah duplikat sampel dari kelas minoritas secara acak hingga mencapai keseimbangan antar kelas dalam dataset. Sedangkan algoritma klasifikasi yang digunakan menggunakan algoritma Feedforward ANN atau Multilayer Perceptron.
Metode hibrida Spearman Rho dan RFE, menghasilkan 130 subset fitur baru dari 135 fitur, 25 faktor non-akademis dan 3 dimensi yang memiliki korelasi signifikan dengan variabel target yang bersumber dari MAIF. MAIF menggabungkan tiga kerangka kerja meliputi The Big Five Personality traits (BFP) untuk mengukur dimensi personality mahasiswa, Family Influence scale (FIS) untuk mengukur dimensi dukungan keluarga, dan Higher Education Service Quality (HEISQUAL) untuk mengukur dimensi lingkungan perguruan tinggi. Tahap handling imbalanced dataset menggunakan metode ROS menghasilkan dataset yang balance antara kelas mayoritas dan minoritas masing – masing sebanyak 248 dataset dengan jumlah total
sebanyak 498 dataset. Tahap ketiga menerapkan algoritma ANN untuk klasifikasi biner kinerja akademik mahasiswa. Model DSI mampu menghasilkan kinerja terbaik berdasarkan stabilitas rerata akurasi model, presisi, recall, dan f1-score lebih
dari 99,50%, rerata nilai kurva AUC-ROC sebesar 1,00. Hasil prediksi pada sejumlah data baru diperoleh nilai tingkat validasi hasil prediksi sebesar 81,25%.
Kata Kunci: Model DSI, academic performance, spearman rho, recursive feature elimination, Random Oversampling, neural networks.

Higher Education regulates the academic predicate of students through the achievement of student academic performance, one of which is the graduation time. Student academic performance is influenced by multiple factors from various aspects or dimensions that are considered. The dimensions that affect academic performance include the personal dimension of students, the family dimension, and the dimension of college service quality. Factors from the three dimensions are measured using a questionnaire called t he Multidimensional Academic Influence Framework (MAIF). Monitoring academic performance is important to prevent academic failure which can hurt education providers and students. Academic performance management helps universities in decision-making and enhancing competitive advantage. It also affects the college's academic policy, curriculum improvement, and pedagogical modification of lecturers. Dimensions Student Intelligence (DSI) model as a binary classification model to classify students' academic performance in one of two different classes using the Multidimensional Academic Influence Framework (MAIF) and Artificial Neural Network (ANN) algorithm. This model is optimized in its modeling procedure through data pre-processing stage with feature selection technique using feature selection and imbalanced dataset handling technique using resampling. The Feature Selection technique is performed using the Spearman Rho and Recursive Feature Elimination (RFE) Hybrid methods to determine features that correlate with student graduate time and are relevant to model accuracy and efficiency. The technique of handling imbalanced datasets is done using the Random Oversampling (ROS) method which adds duplicate samples from minority classes randomly until it reaches a balance between classes in the dataset. While the classification algorithm used uses the Feedforward ANN or Multilayer Perceptron algorithm.
The hybrid method of Spearman Rho and RFE, produces 130 new feature subsets from 135 features, 25 non-academic factors, and 3 dimensions that have significant correlations with target variables sourced from MAIF. MAIF incorporates three frameworks including The Big Five Personality traits (BFP) to measure student personality dimensions, the Family Influence scale (FIS) to measure family support dimensions, and Higher Education Service Quality (HEISQUAL) to measure college environment dimensions. The stage of handling imbalanced datasets using the ROS method produces a balanced dataset between the majority and minority classes of 248 datasets each with a total of 498 datasets. The third stage applies the ANN algorithm for binary classification of student academic performance. The DSI model can produce the best performance based on the stability of the average model accuracy, precision, recall, and f1-score of more than 99.50%, the average AUCROC curve value of 1.00. The prediction results on several new data obtained a prediction validation rate of 81.25%.
Keywords: DSI model, academic performance, spearman rho, recursive feature elimination, random oversampling, neural networks.

Actions (login required)

View Item