Page 240 - Big Data Analytics for Intelligent Healthcare Management

9.2 BACKGROUND 233

considering the probability of the class with respect to a particular attribute by applying Bayes’ the-
orem. This classifier is usually used when users’ ratings or likes vary with respect to the time required
for building the required model [20].

9.2.3.3 Evaluation of recommendation system
A recommendation system should be properly evaluated using different measures. Without evaluation,
we cannot identify whether our proposed recommendation system is correctly recommending and pre-
dicting or not. There are generally two measures that we consider for evaluation that is, prediction
accuracy and precision of the recommendation list. The quality of different filtering-based recommen-
dation systems is evaluated by calculating accuracy and coverage. Accuracy is the ratio of accurate
recommendations to total possible recommendations and coverage can be defined as the percentage
of objects (items) in the domain that the recommendation system is able to recommend [15].
Accuracy metrics are used to evaluate the accuracy of any type of filtering-based recommenda-
tion system by contrasting the predicted ratings directly with the actual user rating. We use different
statistical accuracy metrics such as Mean Absolute Error (MAE) and Root Mean Square Error
(RMSE) etc.
• Mean Absolute Error (MAE)

This measure is very easy to understand and widely used for calculating the amount of diversion of
recommendation from user’s specific value. It is described in Eq. (9.4).
1 X
MAE ¼ j p u,i r u,i j (9.4)
N
u,i
Where N denotes total number of ratings on the item set, p u,I denotes the predicted rating for user u on
the item i and r u,i is the actual rating. The lower the MAE, the better the accuracy of the recommen-
dation system.
• Root Mean Square Error (RMSE)
This measure defines standard deviation of the residual errors that is, differences between predicted
values and known values. RMSE is described in Eq. (9.5).
1 X 2
RMSE ¼ ð p u,i r u,i Þ (9.5)
N
u,i
Where N denotes total number of ratings on the item set, p u,i denotes predicted rating for user u on the
item i and r u,i is the actual rating. The lower the RMSE, the better the accuracy of the recommendation
system.
Recommendation accuracy metrics are used to recommend any product to the user by calculating
measurement factors such as recall, F-measure, and precision etc. These factors are calculated with the
help of a confusion matrix [15]. The confusion matrix is described in Table 9.1.
Given a classifier and an instance, there are four possible outcomes:
1. True positive (TP): If the instance is positive and it is classified as positive
2. False negative (FN): If the instance is positive but it is classified as negative
3. True negative (TN): If the instance is negative and it is classified as negative