MLBA_02

Classification

Responses in categories/classes

Prediction

Numerical value predicted (not class)

Association Rules

Finding Associations/patterns between items in large databases

Collaborative Filtering

Suggests what goes with what from an individual based on user history and measurable behaviors

Data Reduction

Consolidating large number of records into a smaller set

Dimension Reduction

Reducing number of variables

Data Reduction

_____ reduces number of rows in .xlsx file

Dimension Reduction

_____ reduces number of columns in an .xlsx file

Data Visualization

Data exploration by creating charts and dashboards

Data Visualization is also called _____

Visual Analytics

Visual Analytics is also called _____

Data Visualization

What charts are created in numerical visual analytics

Histograms and Boxplots

What charts are created in categorical visual analytics

Bar Charts

Supervised Learning Algorithm

Uses training data to teach model, gives validation data to test how it does compared to other models, and then gives test data to tell how well it will do

Training Data

Data used to teach supervised learning algorithm

Validation Data

Used to test how the supervised learning algorithm compares to other models

Test Data

Tells how well a supervised learning algorithm will do

Does training data have known or unknown outcomes?

Known

Does validation data have known or unknown outcomes?

Known

Does testing data have known or unknown results?

Unknown

Unsupervised Learning Algorithm

Has no outcome to predict or classify, thus no learning necessary

Machine Learning Project Steps

1. Collect data.

2. Explore, clean, and process data.

3. Reduce dimensions, if necessary

4. Determine machine learning task

5. Partition Data (training, validation, test)

6. Choose machine learning techniques to use

7. Use algorithms to perform tasks

8. Interpret results of algorithm

9. Deploy model

SEMMA

Sample, Explore, Modify, Model, Assess

S in Semma

Sample: Take sample from data set, partition in three

E in SEMMA

Explore: Examine data, statistically and geographically

M1 in SEMMA

Modify: Transform variables and impute missing values

M2 in SEMMA

Model: Fit predictive models

A in SEMMA

Assess: Compare using validation data set

Preliminary Steps

1. Organize Data

2. Sample from database

3. Oversample rare events in classification tasks

4. Process and clean data

5. Handle categorical variables

6. Select variables

7. How many variables, how much data?

8. Outliers

9. Missing values

10. Normalize/Standardize, rescale data

Why sample from database?

Models don’t need to be inundated with data to be accurate

Why oversample rare events in classification tasks

If we sample something rare, we may have too much common data to create an accurate model

What are ways of handling categorical variables

Code numerically

How many variables and how much data?

10 records per variable is good rule of thumb

How to identify and what to do with outliers?

3+ std dev from mean, determine if outlier values are wrong, natural, or what is being sought after.

How to handle missing values

Omit records, replace with imputed value (like mean), determine if data is unnecessary or if investment is needed

Normalizing Data

Standardizing data, normalize as z-score to bring variables onto one scale

Data Partitioning

Dividing data into training, validation, and test d ata

Overfitting

Model fits training data too well and thus is ineffective in predicting future outcome values

Using a smaller data set is likely to generate a(n) _____ result, therefore ________________________________

ACCURATE, therefore number of required records fits into rows of an excel spreadsheet