Machine Learning Discussion

Comment: Use machine Learning and faithfully follow what is being asked in the questions,according to the instructions.pdf file. Each question must be answered with a print of the
computer screen. The program to be used is Mathlab. The databases were made available
for the completeness of the work **There is another 60mb “signs” database file that I couldn’t
attach here.
1 – Using the Extracting_characteristics_tempo code and the Signals.mat file. Perform
the procedures below:
a) Load the Signals.mat file
b) Design a low-pass filter of order 20 with a cutoff frequency of 1 kHz, knowing that the
signal was sampled with Fs=50 kHz.
c) Perform the filtering of the channels referring to the vibration signals of the Y and Z
directions of the inner bearing (columns 3 and 4).
d) Extract the characteristics below from the vibration signals obtained in letter c.
e) Build the feature matrix and obtain its dimension.
f) Save the features matrix
2 – Using the Hold-out technique to separate 70% of the data for training 10% for
validation and 20% for testing, design a regression model using the database ship
data according to the items below:
a) Tune the Random Forest algorithm using the Grid Search technique varying the number
of trees by [20,40] with an increment of 2 trees at each iteration, consider the minumum leaf
size equal to 5.
b) Tune the SVM algorithm using the Grid Search technique by varying the term of
regularization [100, 500] with increment of 100 at each iteration, use the kernel function
Gaussian, kernel scale auto, episilon mode auto and perform data standardization.
c) Obtain the values of RMSE, MSE, MAE and R2 in the test set of the best tuned models.
d) Compare the results of the two tuned regressive algorithms and select the best for this
database. Justify your answer.
3 – Using the Hold-out technique to separate 70% of the data for training, 10% for
validation and 20% for testing, design a classification model using the database
ABVT. On this basis, the last column is the class identification. Target 6 corresponds
to imbalance fault, 7 corresponds to horizontal misalignment fault and 8 corresponds
to the vertical misalignment fault.
a) Tune the K-NN algorithm using the Grid Search technique by varying the number of
neighbors of [3,30] with an increment of 3 neighbors at each iteration, use the Euclidean
distance, equal distance weight, and standardize the data.
b) Calculate the average precision, average recall and average F1-score in the test set for
the tuned model obtained in letter a.
c) Obtain the ROC curve of the test set for the classifier tuned to letter a and check which
class was best identified.
4 – Using the SBRT database apply the K-fold technique with 10 folds to separate the
training and validation data and set aside 15% for testing. run the items below.
a) Load the SBRT database, perform the classification using Artificial Neural Network with 2
layers, with 16 neurons in each layer, using the sigmoid activation function, with a minimum
number of iterations equal to 1000 and standardize the data. Get the matrix of confusion for
the test data.
b) Apply the Random undersampling technique to the majority classes and save the base as
base_subsampled. Perform sorting on the subsampled_base using the Artificial Neural
Network configuration of the letter a. Get the confusion matrix for the test data.
c) Apply the Smote technique through the code Criando_instancias_smote in the instances
of minority class training and save the database as training_data. perform the classification
of the training_dataset using the Artificial Neural Network configuration of letter a. Get the
confusion matrix. Load the data_test file and generate the matrix of confusion.
d) Compare the results presented in items a, b and c and check whether the use of Random
undersampling technique and the Smote technique were able to improve the efficiency of the
model in relation to the minority class.
5 – Using the ship database apply the K-fold technique with 5 folds and separate
20% for testing. Perform the items below.
a) Perform the regression using an SVM optimized through the Random Search technique
with 20 iterations. Leave the Quadratic Kernel fixed and standardize the data. The other
parameters of the regressor must be optimized. Obtain the RMSE, MSE, MAE, and R2
values of the algorithm tuned to the test data.
b) Apply the PCA technique to select 3 principal components. Tune the SVM following the
instructions in letter a. Obtain the RMSE, MSE, MAE, and R2 values of the algorithm tuned
to the test data.
c) Apply the mRmR technique and select the 3 most relevant features. Tune the SVM
following the instructions in letter a. Obtain the RMSE, MSE, MAE, and R2 values of the
algorithm tuned to the test data.
d) Apply the RelieFF technique and select the 3 most relevant features. Tune the SVM
following the instructions in letter a. Obtain the RMSE, MSE, MAE, and R2 values of the
algorithm tuned to the test data.
e) Compare the results presented in items a, b, c and d. Check whether the use of
dimensionality reduction techniques were able to improve the model’s effectiveness. If so,
which was the best among those evaluated?