Optimasi k-nearest neighbors menggunakan fuzzy c-means pada ketepatan waktu kelulusan mahasiswa

ABSTRAK:

Ketepatan waktu kelulusan mahasiswa perlu diprediksi karena berpengaruh padakeberhasilan evaluasi akademik. Evaluasi akademik memiliki tantangan dalampelaksanaanya seperti adanya metode yang beragam, pendekatan evaluasi yang berbeda-beda, dan kesulitan mengukur kriteria abstrak. Prediksi ketepatan waktu kelulusanmahasiswa sebagai solusi yang ditawarkan pada penelitian ini untuk mengoptimalkanevaluasi akademik. Objek penelitian ini yaitu mahasiswa Program Studi TeknikInformatika UIN Malang. Dalam kurun waktu 2014 sampai 2018, rerata tingkat kelulusantepat waktu mahasiswa hanya 29%. Angka tersebut tergolong rendah jika dibandingkandengan standar persentase penilaian akreditasi oleh BAN-PT. Tujuan penelitian ini yaitumembuat model prediksi yang dapat melakukan klastering data dan prediksi ketepatanwaktu kelulusan mahasiswa. Penelitian ini menggunakan 2 metode; fuzzy c-means(algoritma klastering data) dan k-nearest neighbors (algoritma prediksi data).Implementasi algoritma memperhitungkan 6 atribut yaitu indeks prestasi semester 1-4,jenis kelamin, jenis pembiayaan. Dataset didapat dari data akademik mahasiswa tahun2014 sampai 2018. Dataset diolah dengan 4 tahap preprocessing; data cleaning, dataintegration, data transformation, data reduction. Uji coba dilakukan dengan komposisidata training dan testing 60:40 dan 70:30. Dalam pengimplementasian algoritma fuzzy c-mean dilakukan skenario kustomisasi nilai variabel c=2, c=3, c=4. Dalampengimplementasian algoritma k-nearest neighbors dilakukan skenario kustomisasi nilaivariabel k=1, k=3, k=5, k=7. Hasil uji coba menunjukkan bahwa dari 16 model prediksiyang dihasilkan, model prediksi 11 memiliki akurasi terbaik yaitu 74.7% dengankomposisi data training dan data testing 60:40, kustomisasi nilai c=3, kustomisasi nilaik=5. Model terbaik yang ditemukan dapat digunakan untuk memprediksi persentasetingkat kelulusan tepat waktu mahasiswa di Program Studi Teknik Informatika. Persentasetersebut dapat digunakan sebagai acuan program studi untuk memperbarui, menambah,atau memperbaiki kebijakan akademik dalam rangka mengoptimalkan evaluasi akademik.

ABSTRACT:

The timeliness of student graduation needs to be predicted because it affects the
success of academic evaluations. Academic evaluation has challenges in its implementation
such as the existence of various methods, different evaluation approaches, and difficulty
measuring abstract criteria. Predicting student graduation timeliness as a solution offered
in this study to optimize academic evaluation. The objects of this research are students of
the UIN Malang Informatics Engineering Study Program. From 2014 to 2018, the average
on-time graduation rate for students was only 29%. This figure is low when compared to
the standard percentage for accreditation assessment by BAN-PT. The purpose of this study
is to create a predictive model that can cluster data and predict the timeliness of student
graduation. This research uses 2 methods; fuzzy c-means (data clustering algorithm) and
k-nearest neighbors (data prediction algorithm). Algorithm implementation takes into
account 6 attributes, namely grade point average 1-4, gender, type of financing. The dataset
was obtained from student academic data from 2014 to 2018. The dataset was processed in
4 stages of preprocessing; data cleaning, data integration, data transformation, data
reduction. The trials were carried out with the composition of training and testing data
60:40 and 70:30. In implementing the fuzzy c-mean algorithm, scenarios for customizing
the values of variables c=2, c=3, c=4 are carried out. In implementing the k-nearest
neighbors algorithm, scenarios for customizing the variable values k=1, k=3, k=5, k=7 are
carried out. The results show that of the 16 prediction models produced, prediction model
11 has the best accuracy, namely 74.7% with a composition of training data and testing
data of 60:40, customizing the value of c = 3, customizing the value of k = 5. The best
model found can be used to predict the percentage of on-time graduation rate of students
in the Informatics Engineering Study Program. This percentage can be used as a reference
for study programs to update, add, or improve academic policies in order to optimize
academic evaluation

مستخلص البحث:

ﻳﺠﺐ ﺗﻮﻗﻊ ﺗﻮﻗﻴﺖ ﺗﺨﺮﻣﺜﻞ وﺟﻮد ﻃﺮقج اﻟﻄﻼب ﻛﺤﻞ ﻣﻘﺪم ﻓﻲﻣﺨﺘﻠﻔﺔ ، وﻣﻘﺎرﺑﺎت ﺗﻘﻴﻴﻢ ﻣﺨﺘﻠﻔﺔ ، وﺻﻌﻮﺑﺔ ﻗﻴﺎس اﻟﻤﻌﺎﻳﻴﺮ اﻟﻤﺠﺮدة. اﻟﺘﻨﺒﺆ ﺑﻤﻮاﻋﻴﺪ ﺗﺨﺮﻫﺬﻩ اﻟﺪراﺳﺔ ﻟﺘﺤﺴﻴﻦ اﻟﺘﻘﻴﻴﻢ اﻷﻛﺎدﻳﻤﻲ.أﻫﺪاف ﻫﺬا اﻟﺒﺤﺚ ﻫﻢ ﻃﻼب ﺑﺮﻧﺎﻣﺞ دراﺳﺔ ﻫﻨﺪﺳﺔ اﻟﻤﻌﻠﻮﻣﺎﺗﻴﺔUIN Malang. ﻣﻦﻋﺎم2014إﻟﻰ ﻋﺎم2018ﻛﺎن ﻣﺘﻮﺳﻂ ،ﻣﻌﺪلاﻟﺘﺨﺮجﻓﻲاﻟﻮﻗﺖاﻟﻤﺤﺪدﻟﻠﻄﻼب29٪ﻓﻘﻂ. ﻫﺬا اﻟﺮﻗﻢ ﻣﻨﺨﻔﺾ ﺑﺎﻟﻤﻘﺎرﻧﺔﻣﻊ اﻟﻨﺴﺒﺔ اﻟﻤﺌﻮﻳﺔ اﻟﻘﻴﺎﺳﻴﺔ ﻟﺘﻘﻴﻴﻢ اﻻﻋﺘﻤﺎد ﻣﻦ ﻗﺒﻞBAN-PT.اﻟﻐﺮض ﻣﻦ ﻫﺬﻩ اﻟﺪراﺳﺔ ﻫﻮ إﻧﺸﺎء ﻧﻤﻮذج ﺗﻨﺒﺆي ﻳﻤﻜﻨﻪ ﺗﺠﻤﻴﻊج اﻟﻄﻼب. ﻳﺴﺘﺨﺪم ﻫﺬا اﻟﺒﺤﺚ ﻃﺮﻳﻘﺘﻴﻦ ؛اﻟﺒﻴﺎﻧﺎت واﻟﺘﻨﺒﺆ ﺑﻤﻮاﻋﻴﺪ ﺗﺨﺮFuzzy C-Means)ﺧﻮارزﻣﻴﺔ ﺗﺠﻤﻴﻊ اﻟﺒﻴﺎﻧﺎت( وK-Nearest Neighbours)ﺧﻮارزﻣﻴﺔ ﺗﻮﻗﻊ اﻟﺒﻴﺎﻧﺎت(. ﻳﺄﺧﺬ ﺗﻄﺒﻴﻖ اﻟﺨﻮارزﻣﻴﺔ ﻓﻲ اﻻﻋﺘﺒﺎر6ﺳﻤﺎت وﻫﻲ اﻟﻤﻌﺪل اﻟﺘﺮاﻛﻤﻲﻓﻲ اﻟﻔﺼﻮل ﻣﻦ1إﻟﻰ4، اﻟﺠﻨﺲ ، ﻧﻮع اﻟﺘﻤﻮﻳﻞ. ﺗﻢ اﻟﺤﺼﻮل ﻋﻠﻰ ﻣﺠﻤﻮﻋﺔ اﻟﺒﻴﺎﻧﺎت ﻣﻦ اﻟﺒﻴﺎﻧﺎت اﻷﻛﺎدﻳﻤﻴﺔ ﻟﻠﻄﻼب ﻣﻦ2014إﻟﻰ2018.ﺗﺘﻢ ﻣﻌﺎﻟﺠﺔ ﻣﺠﻤﻮﻋﺔ اﻟﺒﻴﺎﻧﺎت ﺑﺄرﺑﻊ ﻣﺮاﺣﻞ ﻣﻦ اﻟﻤﻌﺎﻟﺠﺔ اﻟﻤﺴﺒﻘﺔ ؛ ﺗﻨﻈﻴﻒ اﻟﺒﻴﺎﻧﺎت ، ﺗﻜﺎﻣﻞ اﻟﺒﻴﺎﻧﺎت ، ﺗﺤﻮﻳﻞ اﻟﺒﻴﺎﻧﺎت، ﺗﻘﻠﻴﻞ اﻟﺒﻴﺎﻧﺎت. ﺗﻢ إﺟﺮاء اﻟﺘﺠﺎرب ﺑﺎﺳﺘﺨﺪام ﺑﻴﺎﻧﺎت اﻟﺘﺪرﻳﺐ واﻻﺧﺘﺒﺎر60:40و70:30. ﻋﻨﺪ ﺗﻨﻔﻴﺬ ﺧﻮارزﻣﻴﺔFuzzy C-Mean، ﻳﺘﻢ ﺗﻨﻔﻴﺬ ﺳﻴﻨﺎرﻳﻮﻫﺎت ﻟﺘﺨﺼﻴﺺ ﻗﻴﻢ اﻟﻤﺘﻐﻴﺮاتc = 2،c = 3،c = 4. ﻋﻨﺪ ﺗﻨﻔﻴﺬ ﺧﻮارزﻣﻴﺔK-NearestNeighbours، ﻳﺘﻢ ﺗﻨﻔﻴﺬ ﺳﻴﻨﺎرﻳﻮﻫﺎت ﻟﺘﺨﺼﻴﺺ ﻗﻴﻢ اﻟﻤﺘﻐﻴﺮاتk = 1،k = 3،k = 5،k = 7.أﻇﻬﺮت ﻧﺘﺎﺋﺞ اﻻﺧﺘﺒﺎرأﻧﻪ ﻣﻦ ﺑﻴﻦ16ﻧﻤﻮذج ﺗﻨﺒﺆ ﺗﻢ إﻧﺘﺎﺟﻪ ، ﻛﺎن ﻟﻨﻤﻮذج اﻟﺘﻨﺒﺆ11أﻓﻀﻞ دﻗﺔ ﺑﻨﺴﺒﺔ74.68٪ﻣﻊ ﺗﺮﻛﻴﺒﺔ ﺑﻴﺎﻧﺎت اﻟﺘﺪرﻳﺐ وﺑﻴﺎﻧﺎتاﻻﺧﺘﺒﺎر60:40، ﺗﺨﺼﻴﺺ ﻗﻴﻤﺔC = 3، ﺗﺨﺼﻴﺺ ﻗﻴﻤﺔk = 5. ﻳﻤﻜﻦ اﺳﺘﺨﺪام أﻓﻀﻞ ﻧﻤﻮذج ﺗﻢ اﻟﻌﺜﻮر ﻋﻠﻴﻪ ﻟﻠﺘﻨﺒﺆ ﺑﻨﺴﺒﺔج ﻓﻲ اﻟﻮﻗﺖ اﻟﻤﺤﺪد ﻟﻠﻄﻼب ﻓﻲ ﺑﺮﻧﺎﻣﺞ دراﺳﺔ ﻫﻨﺪﺳﺔ اﻟﻤﻌﻠﻮﻣﺎﺗﻴﺔ. ﻳﻤﻜﻦ اﺳﺘﺨﺪام ﻫﺬﻩ اﻟﻨﺴﺒﺔ اﻟﻤﺌﻮﻳﺔ ﻛﻤﺮﺟﻊ ﻟﺒﺮاﻣﺞ اﻟﺪراﺳﺔاﻟﺘﺨﺮﻟﺘﺤﺪﻳﺚ أو إﺿﺎﻓﺔ أو ﺗﺤﺴﻴﻦ اﻟﺴﻴﺎﺳﺎت اﻷﻛﺎدﻳﻤﻴﺔ ﻣﻦ أﺟﻞ ﺗﺤﺴﻴﻦ اﻟﺘﻘﻴﻴﻢ اﻷﻛﺎدﻳﻤﻲ