MachineLearning-Practices

This is a set of practice exercises in Fundamentals of Machine Learning, a subject at HCMUS.

Homework 1: Linear Regression

Notebook link

Dataset relate to how much money a person can get from a mortgage on their home.

In HW1 notebook, I do:

1. Preprocessing data: Replace null values, remove outliers and normalize data
2. Splitting train-test dataset and implement Linear Regression model which consists of basis Linear Reg, Lasso, Ridge, and ElasticNet.
3. Polynomial Transform

All MAE score is less than 0.01

Homework 2: Logistic Regression

Notebook link

Objective's homework is classifying a person having heart disease or not.

Tasks in HW2 is quitely like HW1. I do:

1. Preprocessing data: remove outliers by quantile and normalize data based on feature's distribution
2. Reduction data by dropping feature based on small feature's correlation value
3. Implement Logistic Regression models with different solver and different penalty and compare them by using F1-Score

All F1-Score of classification model is 0.98

Homework 3: Naive Bayes Classification

Notebook link

Using Naive Bayes model to classify the customer buy travel insurance or not.

All tasks is the same above, plus: Using stack learning method for two model NB which is BernoulliNB for categorical features and GaussianNB for numerical features

So, F1-Score of model is 0.72. This score, compared to tradition method, has been improved

Homework 4: K-Means - Unsupervised Learning

Notebook link

This is pratice which about unsupervised learning. Final goal of Homework is to separate dataset to clusters by using K-Means

Applying metrics such as Silhouette, Davies-Bouldin, Calinski-Harabasz for meansuring K-Means's performance:

• Silhouette Score: 0.37
• Davies-Bouldin Index: 1.30
• Calinski-Harabasz Index: 3345.69

Score of three metrics is at an acceptable level.

Homework 5: SVM & MLP

Notebook link

Using SVM and MLP to classify the imbalance dataset (99.4/1.6)

Best F1-Score of SVM is 0.186. And MLP's is 0.183