CS 548 KNOWLEDGE DISCOVERY AND DATA MINING Project 1
Download as PPTX, PDF0 likes209 views
This document summarizes different techniques for handling missing data values, transforming attributes, discretizing continuous variables, evaluating feature subsets using correlation, and performing online analytical processing (OLAP) operations on multidimensional data. These include manually and regression-based imputation of missing values, mapping country names to development scores, equal width and frequency discretization, using the CfsSubsetEval correlation-based algorithm to calculate feature merit, and common OLAP operations like roll-up, slice, and dice.
![Sunrise and sunset[1]](https://cdn.slidesharecdn.com/ss_thumbnails/sunriseandsunset1-100426102718-phpapp02-thumbnail.jpg?width=560&fit=bounds)
CS 548 KNOWLEDGE DISCOVERY AND DATA MINING Project 1
- 1. CS 548 KNOWLEDGE DISCOVERY AND DATA MINING Fall 2016 - Project 1 By: Yousef Fadila ML Tlachac Francisco Guerrero
- 2. Filling in the missing value Discretize: ? = “unknown” Manually filling in the data: ? = Germany GDPPC + Switzerland GDPPC) = 31.35 Regression imputation: GDPPC = 2.1069 * LIFE-EXP + 0.1911 * AC-S-ED + -40.4882 * (SWL= [175-200),[125-150),[200-225), [225-250),[250-275)) -16.6881 *(SWL=[200-225),[225-250),[250-275)) - 100.3841. GDPPC (USA) = 2.1069 * 77.4 + 0.1911 * 94.6 -40.4882 *1 -16.6881 * 1 - 100.3841 = 23.59
- 3. Transforming COUNTRY attribute COUNTRY HDI score COUNTRY HDI score Ethiopia LOW Switzerland VERY-HIGH India MEDIUM Germany VERY-HIGH Mexico HIGH Japan VERY-HIGH Thailand HIGH Canada VERY-HIGH Russia HIGH Brazil HIGH USA VERY-HIGH France VERY-HIGH
- 4. Discretizing AC-S-ED Equal width Equal frequency
- 6. Merit The CfsSubsetEval formula used to calculate merit is ∑corr(aj,t)/√((∑σ(aj)2)+2corr(aj1,aj2)∏σ(aj)) where t is the target attribute (play), and aj are the selected attributes (outlook & humidity). =(corr(outlook,play) + corr(humidity,play))/√(12+12 + 2corr(humidity,outlook)(1)(1)) = (0.1960 + 0.1565)/√(1+1+2 (0.01610)) = 0.3525/√(2.032202) = 0.2473
- 8. Correlation Matrix Remove: numbUrban & medFamIncome
- 9. Multidimensional arrays and OLAP operations Operations: 1. Roll-up time from day to year 2. Slice year == 2014 3. Roll-up patients from individual patients to all
- 10. OLAP operations on car’s sales data 1. Rolling-up 1. Drilling-down 1. Slicing 1. Dicing