Analisa data itu mudah! yang susah adalah menentukan hasilnya itu valid apa tidak, sesuai apa tidak. Mengatasi hal tersebut terpaksa kita harus membandingkan beberapa model yang kita pakai.
Di sini akan dicoba membandingan antara Model Bayesian Linear Regression dengan Neural Network Regression, dengan memakai fasilitas Studio Microsoft Machine Learning. Jadi tidak perlu mebuat coding R satu persatu, tapi bagi yang ingin memperdalam bisa email saya.
Sebelum memulai, tentunya harus punya account di Studio Microsoft Machine Learning, bisa daftar disini. Selanjutnya ikuti langkahnya sbb:
1. Preparation
1.1.Data preparation, use data Boston , from Boston Housing Values in Suburbs of Boston, Source:Belsley D.A., Kuh, E. and Welsch, R.E. (1980) Regression Diagnostics. Identifying Influential Data and Sources of Collinearity. New York: Wiley.
1.2. Use modul Split Data, split data randomly 50% train data and 50% test data
Screen1, Boston raw data
3. Model
3.1. Bayesian Linear Regression
3.2. Neural Network Regression
Screen 2, Flowchart lengkap
4. Train Model
4.1. Use Train Model modul, connect to modul Bayesian, connect to Modul Split Data
4.2. Use Score Model modul, connect right bottom to right bottom Modul Split Data
4.3. Modul Evaluate
5. Execute R Script
5.1. Change the R script in Properties panel for Bayesian Linear Regression
5.2. Connect to Modul Add Rows
Catatan
Di modul R Script, di bagian Properti Panel, harus dimasukkan R script yang baru, lihat di
Screen 3.
Screen 3
Screen 3, Script R yang harus dimasukkan ke Modul Bayesian Linear Regression.
Begitu juga di modul Neural Network Regression, di Panel Properties harus dimasukkan R script baru ada di Screen 4
Screen 4
Screen 4. Script R yang harus dimasukkan ke Modul Neural Net Regression.
6. Neural Network Regression
6.1. Follow the procedure from Preparation to Add Rows, for the Neural Network Regression.
6.2. Connect the Add Rows module Execute R Script in Neural Network Regression.
7. Result.
7.1. Run all application
7.2. Click Visualize table,
7.3. Comparing which is better, Bayesian Linear Regression or Neural Network Regression.
7.4. Check Screen 5.
7.5. Neural Network Regression has RMSE 3.609129(less than RMSE of Bayesia Network Regression ) and Coefficient determination 0.830432 bigger than Bayesian Linear Regression.
Screen 5.
8. Conclusion
8.1. Neural Network Regression better than Bayesian Linear Regression Model.
8.2. Not absolutely true, in the next I will compare this result with model Boosted Decicion Tree Regression, Linear Regression and Decicion Forest Regression.
References
Tidak ada komentar:
Posting Komentar