Upgrade to Pro
— share decks privately, control downloads, hide ads and more …
Speaker Deck
Speaker Deck
PRO
Sign in
Sign up
for free
Predicting irregularities in public bidding: an application of neural networks
Thiago Marzagão
May 28, 2017
Research
0
1.8k
Predicting irregularities in public bidding: an application of neural networks
Thiago Marzagão
May 28, 2017
Tweet
Share
More Decks by Thiago Marzagão
See All by Thiago Marzagão
thiagomarzagao
1
560
thiagomarzagao
0
1.2k
thiagomarzagao
1
1.8k
thiagomarzagao
0
180
thiagomarzagao
0
1.8k
thiagomarzagao
0
77
thiagomarzagao
0
42
Other Decks in Research
See All in Research
shunk031
0
810
noteinc
1
51k
truerichesradio
0
180
rkrokee
0
250
jayagonoy
0
120
yuukit
1
760
tarugoconf
0
1.7k
mihozono
17
10k
kamicup
0
160
yuukit
0
1.2k
mns54
2
540
jriely
0
280
Featured
See All Featured
holman
288
130k
geeforr
332
29k
jlugia
216
16k
ammeep
656
54k
vanstee
117
4.8k
jcasabona
8
550
eitanlees
112
10k
gr2m
83
11k
sachag
446
36k
matthewcrist
73
7.5k
dougneiner
119
7.9k
tmm1
61
9.2k
Transcript
Predicting irregularities in public bidding: an application of neural networks
Observatory of Public Spending
Government contractor doesn’t pay employees Default epidemy in the federal
government: 4 companies went bankrupt Construction company abandons 3 projects Observatory of Public Spending
Observatory of Public Spending what if we could predict which
contractors will become headaches?
Observatory of Public Spending
Observatory of Public Spending impossible to do manually ~25k new
contracts every year
Observatory of Public Spending
Observatory of Public Spending data + neural networks = predictions
Observatory of Public Spending data: - n = 10186 -
9442 (~93%) not problem - 744 (~ 7%) problem - 2011-2016
Observatory of Public Spending data: - Y: has the company
been punished before?
Observatory of Public Spending data: - X: a total of
183 attributes, like: - # of employees - average salary of employees - # of auctions it participated - donated $ to politicians? - …
Observatory of Public Spending neural networks: - two approaches: -
(“traditional”) neural network - deep neural network
Observatory of Public Spending TNN: - 2 hidden layers -
can’t handle 183 attributes - hence must use PCA first
Observatory of Public Spending TNN: - PCA - selected 24
continuous variables based on covariance matrix - PCA reduced 24 variables to 9 components (~70% of variance; all components w/ eigenvalue > 1)
Observatory of Public Spending TNN: - 9 components + 21
binary vars. - 80% training - w/ oversampling - 20% testing - boosting (10 models)
Observatory of Public Spending DNN: - 3 hidden layers -
hundreds of neurons - can handle all 183 variables - can handle complex relationships between the variables
Observatory of Public Spending DNN: - all 183 variables (no
PCA) - no oversampling - 80% training - 20% testing - 5-fold cross-validation
Observatory of Public Spending
Observatory of Public Spending how can we evaluate performance? -
accuracy (% of correct predictions overall) - recall (% of problems predicted to be problems) - precision (% of predicted problems that are problems)
Observatory of Public Spending how can we evaluate performance? -
accuracy (% of correct predictions overall) - recall (% of problems predicted to be problems) - precision (% of predicted problems that are problems)
Observatory of Public Spending results: - TNN precision: 0.24 -
DNN precision: 0.79 - huge difference! extra computational cost of DNN is worth it
Observatory of Public Spending to do: - improve recall -
0.58 w/ TNN - 0.26 w/ DNN - change the law - must allow gov not to contract w/ high risk companies
Observatory of Public Spending Ting Sun tsun9920@gmail.com Leonardo Sales leonado.sales@cgu.gov.br
Observatory of Public Spending @tmarzagao thiagomarzagao.com