Wait A Bike

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 14

The Application Scenario

Data Architecture for the Application Scenario


Implemented Data Value Chain
Analytical Goals and the Chosen Ones

It’s worthwhile wait or


maybe move to
another bench?
Model Implementation
Datasets

➔ Loan_historical
➔ Benches_anchors
➔ Benches
Model Implementation
Data Exploration
➔ 56 benchs
➔ 900 anchors
➔ 1.422 bicycles
➔ More than 1.3M transactions
➔ Data from 2013 to 2017

➔ Missing values
➔ Duplicated loans
➔ No reload information
➔ Useful data from 2015 to 2017
Model Implementation
Data Transformations
Model Implementation
Algorithms and Methods anchor

loan_historical min granularity group by

time
loans
concat add variation column split
putbacks

sort by date max_anchors bench_tracking

inference bikes_reconstruction
Model Implementation
regressor
Splits, Parameter Using, Cross Validation, etc.
hyperparameters features
+65k samples 25 folds 0.001 learning rate lag 2 (14) hour quarter year dayofmonth

672 test set 10k boosting rounds lag 1 (7) lag 3 (21) day of week month dayofyear week of year
Results
Evaluation of Results (metrics)

● RMSE Root Mean Squared Error


● Training and validation data
● Early stopping of 2000
Results
● Even with the assumptions, the results
are promising for creating a
recommendation system.
● The assumption makes the model less
accurate with real data.
Use of Large Language Models
● Data base visualization
● Concepts understanding
● Specific support coding
● Frameworks recommendation
Conclusions
● It is possible to create a benches recommendation system, but it is necessary
to improve the veracity of the data.
● Decision making could be improved if combined with more valuable data for
dates, such as holidays, weather behavior, election days, etc.

You might also like