Machine Learning Startup File.ReadMe

Version	Date	Author	Notes
0.1	19 July 2023	Ken Dizon	Initial version
0.2	26 July 2023	Ken Dizon	Refined all ML categories
1.0	28 July 2023	Ken Dizon	Alpha phase

Objective: write a sample ML script for initial start up. ML gives computers the ability to learn without being explicitly programmed.

CONTENT

• Libraries
• Types of Machine Learning
  • Use Cases
  • Algorithms
• Models & Math
• Startup sample content

---

TYPES: Supervised vs Unsupervised

• Supervised
  • Regression: predicts trends using labeled data
  • Classification: classifies labeled data
• Unsupervised
  • Clustering: finds patterns and groupings from unlabeled data

ML

Data > Feature Extraction > ML Model > Outcome

ML Techniques	Types	Case Prediction
Regression	Simple, Multiple	continuos numerical value
Classification	Binary, Multi-class	class label for unlabled test case
Clustering	Partitioned, Hierarchial, Density	grouping data points by similarity

Use Cases

Regression	Classification	Clustering
Numeric prediction, estimation, forecasting	detection, retension, daignostic	segmentation, target market, recommender systems

Algorithms

Regression	Classification	Clustering
ordinal, poisson, linear, polynomial, lasso, bayesian, NNR, decision forest R, KNNR	DT, naive bayes, Linear discriminant analysis, KNN, logistic regression, NN, SVM	K-means, k-Median, Fuzzy c-Means, agglomerative, divisive, DBSCAN

---

Models & Math

Regression Model	Evaluation	Selection
Linear Regression	MAE, MSE, R2, RSS, Variance Score	Influence of Xs to predict a continuous numeric Y value
Decision Tree Regression [DTR]	mse, mae accuracy	...
Multiclass Prediction	...	...

Classification Model	Evaluation	Selection
DT	DT classification Accuracy	Categorical identification based on characteristics
KNN	mean acc, std acc	...
Logistic Regression	Jaccard index, Confusion matrix (F1-score), logloss	predict probability of categorical dependent variable
SVM	...	Image recognition, text category assignment, detecting spam, sentiment analysis, gene expression

Clustering Model	Evaluation	Selection
k-means	non-overlaps to minimize 'intra' & max 'inter' cluster distance	relatively efficient (med/large data)
...	...	trees of clusters
...	...	arbitrary shaped clusters

---

Machine Learning TYPE Model Name Startup File

Content

• Libraries

1. Load Data
   • 1.1 Data Exploration
2. Data Preprocessing
   • 2.1 Cleaning
   • 2.2 Missing Data
   • 2.3 Scaling
   • 2.4 Feature Engineering/Selection
3. Split - Test & Train
4. Model Selection
5. Model Training
6. Model Evaluation