Fundamentals_of_Machine_Learning

Fundamentals of Machine Learning, circa September 2023, taught by Dr Chico Camargo

CONTENT

Part A - Exploratory Data Analysis and Data Visualisation

• A.1 - Data Preparation
• A.2 - Scatterplot
• A.3:5 - Histograms
• A.6 - Bar Chart

Part B - Training Classifiers

• B.7 - Data Preparation
• B.8 - Train-Test Split
• B.9 - Classification using a Perceptron (Confusion Matrix)
• B.10 - Classification using Logistic Regression (Confusion Matrix)
• B.11 - Importance of Train-Test Split
• B.12 - Perceptron vs Logistic Regression
• B.13 - Changeable data or not?

Part C - Linear Regression

• C.14 - Data Preparation and Visualisation (Scatterplot (4))
• C.15 - Pearson Correlation
• C.16 - Linear Regression
• C.17 - K-Fold Cross-Validation
• C.18 - More Columns = Higher R2 Score?
• C.19 - Preferred Linear Regression Model and Classifier Accuracy

Part D - Clustering

• D.20 - Data Visualisation and Preparation (K-Means Clustering)
• D.21 - DBSCAN Clustering
• D.22 - Silhouette Score between K-Means and DBSCAN
• D.23 - Davies-Bouldin Score between K-Means and DBSCAN
• D.24 - Agglomerative Clustering
• D.25 - Why does DBSCAN run later than K-Means? (eps parameter)
• D.26 - Scenario where DBSCAN is better than K-Means and vice versa

Part E - Model Selection

• E.27 - Comparison of models (testing error curve)
• E.28 - Comparison of models (bias-variance trade-off)
• E.29 - Overfitting and its issues

Part F - Dimensionality Reduction

• F.30 - Advantages and Disadvantages of using PCA, t-SNE, and UMAP for Visualisation
• F.31 - When to use PCA, t-SNE, and UMAP
• F.32 - Which Dimensionality Reduction Technique for Interpretability and Stability? (Stochasticity of t-SNE and UMAP)

Part G - Applications of Machine Learning

• G.33 - Scenario where High Accuracy Rate is Invalid (Medical Diagnosis)
• G.34 - Achievability of 100% precision and 100% recall in classification
• G.35 - High Precision and Low Recall or High Recall and Low Precision (Spam Filter and Fraud Detection)
• G.36 - Scenarios where Machine Learning Algorithm achieves high accuracy or low accuracy (Image Classification)
• G.37 - Scenarios where Machine Learning Algorithm achieves low error but anomaly (Facial Recognition)