Estimasi / Regresi
Aldi Pratama
2
Numerik, Numerik Forecasting Linear Regression (LR), Neural Network (NN), Deep Learning (DL),
ada time series Support Vector Machine (SVM), Generalized Linear Model (GLM),
Ex: harga saham, nilai tukar, inflasi, cuaca, dll
Numerik dan Nominal Classification Decision tree, K-nearest neighbor (KNN), linier discriminant analysis
nominal (LDA), logistic regression (LogR)
Ex: prediksi perilaku konsumen, sentiment analysis, prediksi
kebrankutan, prediksi kepatuhan wajib pajak
Numerik dan N/A Clustering K-Means, Fuzzy C-Means, Self-Organizing Map (SOM), K-Medoids
nominal
Numerik dan N/A Asscociation FP-Growth, A Priori, Coefficient of Correlation, Chi Square, etc
nominal
Peran data analytics (data mining) 3
SUPERVISED LEARNING
• Pembelajaran dengan guru, data set memiliki
target/label/class
• Sebagian besar algoritma data mining
(estimation, prediction/forecasting,
classification) adalah supervised learning
Estimasi Prediksi Klasifikasi
(estimation) (forecasting) (classification) • Algoritma melakukan proses belajar
berdasarkan nilai dari variabel target yang
terasosiasi dengan nilai dari variable
prediktor
5
Machine
Learning vs
Statistics
6
Contoh Penerapan 7
ANALISIS REGRESI
It doesn't account for the It does account for positive or It does account for positive or It does account for positive
direction of the value. Even if negative value. negative value. or negative value.
value is negative, positive value
is used for calculation.
RMSE & MSE share many RMSE & MSE share many
properties with MSE because properties with MSE because RMSE
RMSE is simply the square root is simply the square root of MSE.
of MSE.
MAE is less biased for higher MSE is highly biased for higher RMSE is better in terms of
values. It may not adequately values. reflecting performance when
reflect the performance when dealing with large error values.
dealing with large error values
RMSE is more useful when lower
residual values are preferred
MAE is less than RMSE RMSE tends to be higher than MAE
as the sample size goes up as the sample size goes up.
Evaluation Metrics 13
MAE doesn’t necessarily MSE penalize large errors. RMSE penalize large errors. RMSLE doesn’t penalize large errors.
penalize large errors. It is usually used when you don't
want to influence the results if there
are large errors. RMSLE penalize
lower errors
MAE is more useful when the RMSE is more useful when the
overall impact is proportionate overall impact is disproportionate
to the actual increase in error to the actual increase in error
For example- if error values go For example- if error values go up
up to 6 from 3, actual impact on to 6 from 3, actual impact on the
the result is twice result is more
than twice
When actual and predicted When actual and predicted values
values are low, RMSE & are low, RMSE &
RMSLE are usually same RMSLE are usually same
When either of actual or When either of actual or predicted
predicted values are high, RMSE > values are high, RMSE >
RMSLE. RMSLE.
Analisis Regresi 14
Fenomena
Analisis Regresi 15
RMSE = 3.449895507408725.
R2= 0.9830071790386679
Bias vs Varians 17
19