Data Science

Question 1

Recall

Answer 1

True positive / (True positive + False negative)

Question 2

Precision

Answer 2

TP / (TP + FP)

Question 3

F1

Answer 3

2(recall x precision) / (recall + precision)

Question 4

Accuracy

Answer 4

(TP + TN) / (TP + TN + FP + FN)

Question 5

Error

Answer 5

1-Accuracy

Question 6

WCSS F (K Means)

Answer 6

The sum of the sum of distances of each point in each cluster

Sum(Sum(Distance of point to cluster,Points in cluster),Clusters)

Question 7

Centroid

Answer 7

Sum of points / number of points

Question 8

Silhouette coefficient meaning

Answer 8

Close to 1 means good clustering
Close to 0 means that object can belong to either cluster
Close to -1 means bad clustering

Question 9

Silhouette coef calc

Answer 9

A = sum(dist(point to all other points in cluster))/ number of other points in cluster

M = sum(dist( point to all points in other cluster)) / num points in other cluster

N = sum(dist( point to all points in other cluster)) / num points in other cluster

B = min (M, N)

SC = (b-a)/max(a,b)

Question 10

Support
X -> Y

Answer 10

Number of transactions with X / total transactions

Question 11

Confidence
X -> Y

Answer 11

Total number of transactions from support with Y / total number of support transactions

Question 12

Agglomerative

Answer 12

Each object is cluster

two clusters that are the closest. Distance between two clusters C1 and C2 is defined as min distance between object O1 in C1 and O2 in C2

Repeat above step until user specified condition is met (k-clusters)

Question 13

True Pos Rate

Answer 13

TP / (TP+FN) = TP / P

Question 14

True Neg Rate

Answer 14

TN / (TN + FP) = TN / N

Question 15

False Pos Rate

Answer 15

FP / (TN + FP) = FP / N

Question 16

False Neg Rate

Answer 16

FN / (TP + FN) = FN / P

Question 17

KMeans

Answer 17

Cluster objects based on cluster centers

Object belongs to cluster with nearest centroid

Question 18

Purity

Answer 18

Clusters assigned class based on most frequent class in cluster

Purity is number of correctly assigned objects / total objects

1 is perfect, 0 is bad

Question 19

SVM

Answer 19

Supervised machine learning algo

Find line that distinctively classifies data points

Question 20

Apriori Algo

Answer 20

Given a Min confidence and / or support, find all association rules

Any subset of a frequent itemset must be frequent

Any superset of a non-frequent itemset must not be frequent

find sets of otems that have min support starting fro 1-itemsets and expanding to k-itemsets; if a j-itemset is already not frequent then do not consider any superset of it