國立虎尾科技大學 |

Decision Thresholding in Classification-Based Recommendation Systems.

紀錄類型:	書目-語言資料,手稿 : Monograph/item
正題名/作者:	Decision Thresholding in Classification-Based Recommendation Systems./
作者:	Burkman, John Bradford.
面頁冊數:	1 online resource (105 pages)
附註:	Source: Dissertations Abstracts International, Volume: 85-12, Section: B.
Contained By:	Dissertations Abstracts International85-12B.
標題:	Computer science. -
電子資源:	click for full text (PQDT)
ISBN:	9798383179611

Decision Thresholding in Classification-Based Recommendation Systems.
Burkman, John Bradford.

Decision Thresholding in Classification-Based Recommendation Systems. - 1 online resource (105 pages)

Source: Dissertations Abstracts International, Volume: 85-12, Section: B.

Thesis (Ph.D.)--University of Louisiana at Lafayette, 2024.

Includes bibliographical references

Artificial Intelligence (AI) binary recommendation systems input a feature vector and use a model learned on historical data to return a prediction, p ∈ [0,1]. If p > θ for the decision threshold θ, then the recommendation is "Yes"; otherwise, if p < θ, the recommendation is "No." The modeling process itself does not choose the optimal value for θ; such a choice is application-dependent and often involves value judgements and balancing priorities.This work explores methods for selecting a decision threshold in a scenario with imbalanced costs. New cell phones can detect the deceleration profile of a car crash and automatically notify police. When should an ambulance be dispatched based solely on such an automated notification? In this application, the costs are inherently imbalanced. Saving lives takes priority over saving money, but striking the right balance is a complex political and ethical decision, not a purely technical one. I propose three ways to choose the decision threshold to balance the fiscal and ethical quandary, based on (1) total increase in costs, (2) total success rate (precision), and (3) marginal probability that the ambulance is needed. For each criterion I define a metric to calculate the corresponding decision threshold, then compare the models on recall, the proportion of needed ambulances immediately dispatched.The quality of the recommendations depends heavily on the features of the input data, but some kinds of information are more accessible than others. I consider three categories of features and show that the recall increases significantly with better information.This work makes three original contributions to the field. First, I develop ways to choose a decision threshold to balance ethical and fiscal goals, which I have not see in the crash analysis literature. Second, I impute missing data in a public dataset (CRSS), expanding on the methods used by the authors of the dataset and comparing with other methods. Third, given the emerging technology of automated crash notifications from cell phones, I explore whether we can build an AI recommendation system to immediately dispatch ambulances to some crashes but not others in a way consistent with our values.

Electronic reproduction.
Ann Arbor, Mich. :
ProQuest,
2024

Mode of access: World Wide Web

ISBN: 9798383179611Subjects--Topical Terms:

573171
Computer science.
Subjects--Index Terms:

Ambulance dispatchIndex Terms--Genre/Form:

554714
Electronic books.

Decision Thresholding in Classification-Based Recommendation Systems.
LDR:03686ntm a22004217 4500 001 1150193
005 20241022111611.5
006 m o d
007 cr bn ---uuuuu
008 250605s2024 xx obm 000 0 eng d
020 $a 9798383179611
035 $a (MiAaPQ)AAI31292652
035 $a AAI31292652
040 $a MiAaPQ $b eng $c MiAaPQ $d NTU
100 1 $a Burkman, John Bradford. $3 1476634
245 1 0 $a Decision Thresholding in Classification-Based Recommendation Systems.
264 0 $c 2024
300 $a 1 online resource (105 pages)
336 $a text $b txt $2 rdacontent
337 $a computer $b c $2 rdamedia
338 $a online resource $b cr $2 rdacarrier
500 $a Source: Dissertations Abstracts International, Volume: 85-12, Section: B.
500 $a Advisor: Chu, Chee-Hung Henry.
502 $a Thesis (Ph.D.)--University of Louisiana at Lafayette, 2024.
504 $a Includes bibliographical references
520 $a Artificial Intelligence (AI) binary recommendation systems input a feature vector and use a model learned on historical data to return a prediction, p ∈ [0,1]. If p > θ for the decision threshold θ, then the recommendation is "Yes"; otherwise, if p < θ, the recommendation is "No." The modeling process itself does not choose the optimal value for θ; such a choice is application-dependent and often involves value judgements and balancing priorities.This work explores methods for selecting a decision threshold in a scenario with imbalanced costs. New cell phones can detect the deceleration profile of a car crash and automatically notify police. When should an ambulance be dispatched based solely on such an automated notification? In this application, the costs are inherently imbalanced. Saving lives takes priority over saving money, but striking the right balance is a complex political and ethical decision, not a purely technical one. I propose three ways to choose the decision threshold to balance the fiscal and ethical quandary, based on (1) total increase in costs, (2) total success rate (precision), and (3) marginal probability that the ambulance is needed. For each criterion I define a metric to calculate the corresponding decision threshold, then compare the models on recall, the proportion of needed ambulances immediately dispatched.The quality of the recommendations depends heavily on the features of the input data, but some kinds of information are more accessible than others. I consider three categories of features and show that the recall increases significantly with better information.This work makes three original contributions to the field. First, I develop ways to choose a decision threshold to balance ethical and fiscal goals, which I have not see in the crash analysis literature. Second, I impute missing data in a public dataset (CRSS), expanding on the methods used by the authors of the dataset and comparing with other methods. Third, given the emerging technology of automated crash notifications from cell phones, I explore whether we can build an AI recommendation system to immediately dispatch ambulances to some crashes but not others in a way consistent with our values.
533 $a Electronic reproduction. $b Ann Arbor, Mich. : $c ProQuest, $d 2024
538 $a Mode of access: World Wide Web
650 4 $a Computer science. $3 573171
650 4 $a Engineering. $3 561152
650 4 $a Information technology. $3 559429
653 $a Ambulance dispatch
653 $a Decision threshold
653 $a Emergency medical services
653 $a Imbalanced costs
653 $a Imbalanced data
653 $a Machine learning
655 7 $a Electronic books. $2 local $3 554714
690 $a 0984
690 $a 0489
690 $a 0800
690 $a 0537
710 2 $a ProQuest Information and Learning Co. $3 1178819
710 2 $a University of Louisiana at Lafayette. $b Computer Science. $3 1195497
773 0 $t Dissertations Abstracts International $g 85-12B.
856 4 0 $u http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=31292652 $z click for full text (PQDT)