Prediksi pembelotan konsumen software antivirus "X" dengan binary logistic regression dan logistic regression ensembles

Asfihani, Ayu (2015) Prediksi pembelotan konsumen software antivirus "X" dengan binary logistic regression dan logistic regression ensembles. Undergraduate thesis, Institut Teknologi Sepuluh Nopember.

[thumbnail of 1311100084-Undergraduate.pdf]
Preview
Text
1311100084-Undergraduate.pdf

Download (3MB) | Preview

Abstract

Penelitian pembelotan konsumen Perusahaan ‘X’ sangat dibutuhkan ka-rena mengukur loyalitas konsumen tidaklah mudah, hal ini terutama disebabkan oleh Big Data dan kelas pembelotan konsumen yang tidak seimbang. Setelah se-belumnya diteliti oleh Prasasti, dkk (2013), Prasasti dan Ohwada (2014), dan Martono dkk (2014) dengan Machine Learning, pembelotan konsumen Perusa-haan ‘X’ diteliti dengan Binary Logistic Regression dan LORENS untuk mem-bentuk model hingga melakukan pengklasifikasian. Dengan 500000 konsumen Low Price, 408810 konsumen Medium Price, dan 709899 konsumen High Price, hubungan antara variabel Akumulasi Update, Harga Produk, Jawaban Kontrak, Tipe Konsumen, dan Status Pengiriman dengan Pembelotan Konsumen dijelas-kan oleh Binary Logistic Regression. Namun metode tersebut tidak mampu me-nangani Big Data dalam tahap pengujian parameter. Harga Produk dan Tipe Konsumen terbukti berpengaruh signifikan terhadap model namun tidak dapat menjelaskan kecenderungan pembelotan konsumen Medium Price dan High Pri-ce. Karena itu, hasil analisis dan klasifikasi dengan Binary Logistic Regression dalam penelitian ini kurang dapat dipercaya. Dengan kemampuan menangani Big Data, ketidak-seimbangan respon, serta ketimpangan jumlah prediktor ter-hadap banyak pengamatan, LORENS digunakan. Meskipun begitu, LORENS ti-dak mampu menjelaskan hubungan antar variabel karena tidak dapat mengha-silkan model yang interpretatif. Meskipun akurasi klasifikasi Binary Logistic Re-gression terlihat sedikit lebih baik di konsumen Low Price dan Medium Price daripada LORENS (Low Price 66,54 % : 66,25%, Medium Price 77,32% : 74,06 %, High Price 68,42% : 69,04%), metode yang disarankan untuk diaplikasikan untuk data ini adalah LORENS karena lebih dapat dipercaya. Cross Validation yang disarankan LORENS hanya mampu meningkatkan performa LORENS pa-da konsumen High Price dengan akurasi 69,21%.

==========================================================================================================

This research about customer’s defection of ‘X’ Company is really need-ed because measuring their loyalties is not easy, especially because of Big Data and imbalanced customer’s defection responses. After being researched by Prasasti, dkk (2013), Prasasti dan Ohwada (2014), and Martono dkk (2014) with Machine Learning, in this research it will be treated with Binary Logistic Regression and LORENS to form model until perform classification. With 5000 00 Low Price customers, 408810 Medium Price customers, and 709899 High Price customers, relation between variables of Akumulasi Update, Harga Pro-duk, Jawaban Kontrak, Tipe Konsumen, and Status Pengiriman to Pembelotan Konsumen is explained by Binary Logistic Regression. But this method can’t handle Big Data in parameter’s significance testing. Harga Produk and Tipe Konsumen are proven significantly effecting to model but can’t explain any pro-pensity of Medium Price and High Price customer’s defection. Thus, analysis and classification result from Binary Logistic Regression in this research is less trustworthy. With the ability to handle Big Data, imbalanced responses, and un-fairness of predictor’s number to observation’s amounts, LORENS is applied. Nonetheless, LORENS can’t explain any relation between variables because it doesn’t produce any interpretative model. Although classification accuracy from Binary Logistic Regression seems little bit better in Low Price and Medium Pri-ce customers than in LORENS (Low Price 66,54 % : 66, 25%, Medium Price 77, 32% : 74, 06%, High Price 68,42% : 69,04%), the suggested method to be app-lied in this case is LORENS because it is more trustworthy. Cross Validation, as suggested evaluation method by LORENS only able to enhance LORENS’s per-formance in High Price customers with accuracy 69,21%.

Item Type: Thesis (Undergraduate)
Additional Information: RSSt 519.536 Asf p
Uncontrolled Keywords: Big Data; Binary Logistic Regression; Cross Validation; Holdout; Klasifikasi; LORENS
Subjects: Q Science > QA Mathematics > QA278.2 Regression Analysis. Logistic regression
Divisions: Faculty of Mathematics, Computation, and Data Science > Statistics > 49201-(S1) Undergraduate Thesis
Depositing User: - Taufiq Rahmanu
Date Deposited: 24 Jun 2019 06:26
Last Modified: 24 Jun 2019 06:26
URI: http://repository.its.ac.id/id/eprint/63194

Actions (login required)

View Item View Item