China
Africa
Explore chapters and articles related to this topic
I-LED InterventionsPictures, Words and a Stop, Look, Listen
Published in Justin R.E. Saward, Neville A. Stanton, Individual Latent Error Detection (I-LED), 2018
Justin R.E. Saward, Neville A. Stanton
Matthews coefficient (phi) coefficient (Matthews, 1975) can be calculated from the binary values recorded in the contingency table using following equation for phi (Φ): Φ=(TP×TN)−(FP×FN)√(TP+FP)(TP+FN)(TN+FP)(TN+FN) A coefficient of +1 represents perfect positive correlation whereby the I-LED intervention led to all past errors being detected. A coefficient of 0 represents no correlation, whereas a coefficient of −1 indicates a perfect negative correlation.
Intrusion Detection of SCADA System Using Machine Learning Techniques
Published in Sudhir Kumar Sharma, Bharat Bhushan, Narayan C. Debnath, IoT Security Paradigms and Applications, 2020
Mathew’s Correlation Coefficient (MCC): Among all the metrics selected, this parameter as represented by Equation 7.7 in Section 7.7 is the best so far to measure the effectiveness of classification. The phi coefficient in statistics has been renamed to MCC in ML context.
Role requirements in academic recruitment for Construction and Engineering
Published in European Journal of Engineering Education, 2021
Nick Pilcher, Laurent Galbrun, Nigel Craig, Mike Murray, Alan M. Forster, Stuart Tennant
Tables 5–7 show the statistically significant associations within job, research and teaching attributes in ranking order (from highest to lowest number of non-independent attributes). These tables present the key ranking data on the left and more detailed statistical data on the right in line with how data is presented for this specific test and also for others for possible comparative or replicative purposes. Chi-square independence test results (χ2 and p) and the phi coefficient (ø) are given for each non-independent attribute listed in these tables (italic: p < 0.01; non-italic: p < 0.05). Table 5 shows a high number of associations between many of these job attributes. The only attributes with a strong association (i.e ø > 0.7, excluding essential vs. desirable of the same attribute) are PgCE with FHEA (ø = 0.834) and PgCEd with FHEAd (ø = 0.730), meaning that, for example, adverts mentioning a PgCE are highly likely to also mention FHEA. Non-independent attributes with lower phi coefficient values (e.g. ø < 0.4) are still associated, but the associations are weaker. Particularly striking is the lack of any statistical association with Teaching Experience, which could be determined as adverts being ‘biased’ towards research-based attributes with a lack of focus on what could arguably be more important from a student perspective, i.e. teaching-related. Indeed, any shift towards TEF and the ever-increasing focus on National Student Satisfaction Survey (NSS) remains imperceptible, with associations between the teaching-related deemed inconsequential. Drawing on the statistical evidence presented, teaching experience is not paramount when advertising certain ‘academic job roles’.
Models of cruise ports governance
Published in Maritime Policy & Management, 2019
Athanasios A. Pallis, Kleopatra P. Arapi, Aimilia A. Papachristou
Phi (φ) coefficient forms a non-parametric test of relationships that operates on two dichotomous variables; it measures the strength and direction of the association between two variables, while it can be computationally defined as the square root of the ratio of chi-square to the sample size. Similar to a parametric correlation coefficient, the possible values of a Phi coefficient range from −1 to 0 to +1.
Line side placement for shorter assembly line worker paths
Published in IISE Transactions, 2020
To test this hypothesis, we classify the remaining instances by . The resulting partition is depicted in a box plot in Figure 11. Association between two binary variables is commonly measured by the phi coefficient; value indicates a clear relationship between and . For , median is 2%; whereas for , median is 21.5%. Moreover, the upper quantile is 10% for instances, and the lower quantile for instances is 9.5%. Hence, the criterion finds a suitable partition, for which the median values are clearly in the right range and moreover, the quartiles meet at a similar values. Of course, there is an error: of the instances are missed by the criterion, wrongly classified as although they actually have . This error decreases to unrecognized instances. On the other hand, selecting for allows the correct recognition of of the instances with , and of the instances with .