![](bg9.png)
4.8 CRISP-DM Deployment
Once the model is ready and its performance is ac-
cepted, the final phase of the CRISP-DM methodol-
ogy is to deploy it. In most instances, software plat-
forms offer hosting services in the cloud (John et al.,
2021).
In other circumstances, a company might need to
host their in-house software, managing the entire so-
lution and its operation.
The recommendation is to follow existing guide-
lines and policies to safeguard the information and
properly operate and manage the model and its so-
lutions. Popular serverless services include Amazon,
Microsoft, and Google. (Elger and Shanaghy, 2020).
Analysing some popular approaches by the Kag-
gle community, it was evident that focusing on a sta-
tion and not on the complete universe of data yielded
better results that were easy to consume, process,
and manage. Amid the various available models, the
regression family of algorithms consumes fewer re-
sources while producing accurate predictions that suit
the problem perfectly.
An elegant solution that relies on clearly delim-
ited and manageable inputs is generally a better op-
tion than a general solution that does not consider sub-
tle nuances in the items that conform to it. Business
questions can propose scope and applicability prob-
lems, as a product’s core processes supposedly could
provide similar results across many industry and cus-
tomer use cases.
The reality is that understanding and focusing on a
particular problem and offering a solution that best fits
the user’s workflow will undeniably be more success-
ful in the short term. Once the initial problem stops
being a concern, it is safe to switch focus to maintain,
extend, and scale the solution.
5 CONCLUSIONS
California’s groundwater ecosystem does not have
clear seasonality nor easily recognizable trends except
for diminishing water levels. The observations made
to individual stations did allow for accurate predic-
tions to locate water based on area.
With the available results, a secondary study to
map the regional water reservoirs can technically ad-
vise on a better water collection and distribution sys-
tem that uses natural deposits as reserves and as part
of the main supply stream.
The analysis showed that the data’s regularity al-
lowed multiple experiments to allocate just enough
computational power and mathematical framework to
yield an optimal answer that addresses the business
question.
Additional questions regarding the water’s quality,
the effects of continuous drainage, and the amount re-
quired to serve the population, among others, could
be the continuation of the exercise.
A more general view and a broader spectrum of
AI techniques and methods were applied to this multi-
factorial problem to properly analyse and develop the
optimal model. The problem’s characteristics, qual-
ities, and data allowed simple algorithms to gener-
ate highly accurate and effective results. Their out-
put provided high-confidence predictions that satis-
fied the use case from business viability and technical
feasibility perspectives. The next step is to analyse
hidden behaviours and investigate what external fac-
tors can produce additional insights.
REFERENCES
Czako, Z., Sebestyen, G., and Hangan, A. (2021).
Automaticai–a hybrid approach for automatic arti-
ficial intelligence algorithm selection and hyperpa-
rameter tuning. Expert Systems with Applications,
182:115225.
Doremus, H. and Hanemann, M. (2008). The challenges
of dynamic water management in the american west.
UCLA J. Envtl. L. & Pol’y, 26:55.
Elger, P. and Shanaghy, E. (2020). AI as a Service: Server-
less machine learning with AWS. Manning Publica-
tions.
Harou, J. J., Medell
´
ın-Azuara, J., Zhu, T., Tanaka, S. K.,
Lund, J. R., Stine, S., Olivares, M. A., and Jenkins,
M. W. (2010). Economic consequences of optimized
water management for a prolonged, severe drought in
california. Water Resources Research, 46(5).
Ho, S. L. and Xie, M. (1998). The use of arima models
for reliability forecasting and analysis. Computers &
industrial engineering, 35(1-2):213–216.
Inc, A. (2021). 5 types of regression analysis and when to
use them.
John, M. M., Holmstr
¨
om Olsson, H., and Bosch, J. (2021).
Architecting ai deployment: a systematic review of
state-of-the-art and state-of-practice literature. In Soft-
ware Business: 11th International Conference, IC-
SOB 2020, Karlskrona, Sweden, November 16–18,
2020, Proceedings 11, pages 14–29. Springer.
Kadiyala, A. and Kumar, A. (2017). Applications of
python to evaluate environmental data science prob-
lems. Environmental Progress & Sustainable Energy,
36(6):1580–1586.
Li, Y., Zhu, Z., Kong, D., Han, H., and Zhao, Y. (2019). Ea-
lstm: Evolutionary attention-based lstm for time series
prediction. Knowledge-Based Systems, 181:104785.
Negash, K., Khan, B., and Yohannes, E. (2016). Arti-
ficial intelligence versus conventional mathematical
ICSBT 2023 - 20th International Conference on Smart Business Technologies
184