WEBVTT
1
00:00:10.310 --> 00:00:14.349
Hello data standard community. I'm your
host, Shane Waldon, and this is
2
00:00:14.390 --> 00:00:18.149
the data standard audio experience. We
have a very exciting guest today on our
3
00:00:18.190 --> 00:00:23.100
podcast. We have Nelson Saffo,
a digital business intelligence analyst, but the
4
00:00:23.219 --> 00:00:28.179
Atlanta Journal Constitution at Cok Center.
Prise Cock drives new ways of innovation in
5
00:00:28.260 --> 00:00:33.340
the communication and animated industries, and
Nelson has specialties and machine learning, big
6
00:00:33.380 --> 00:00:37.810
day analytics and etl pipelines. Welcome
to the show, Nelson. Thank you
7
00:00:38.009 --> 00:00:43.929
for the good introduction machine. It's
a pleasure talking with you at this point
8
00:00:43.969 --> 00:00:47.570
in time. Thank you. I
am so excited to get another better and
9
00:00:47.689 --> 00:00:50.840
I start off. Can you please
just introduce yourself? Sure, it's a
10
00:00:50.880 --> 00:00:54.719
pleasure talking with you at this point
in time. Shan Yea and the first
11
00:00:54.759 --> 00:00:58.560
I just wanted to clarify. I'm
no longer we talk. has been a
12
00:00:58.679 --> 00:01:02.320
month already, as I look forward
to an advancement in Korea, but of
13
00:01:02.439 --> 00:01:07.189
course, my experience in the mid
of cogs is very happy to share with
14
00:01:07.310 --> 00:01:11.109
you what I've been doing, especially
as this podcast is prior to eat.
15
00:01:11.790 --> 00:01:15.670
So I will just give a brief
some areas an introduction, as the one
16
00:01:15.750 --> 00:01:21.219
sentence description of me, who is
this data study. Individual with takes pleasure
17
00:01:21.379 --> 00:01:26.780
in reaching the gap between customers and
products, with insight for this aision making
18
00:01:26.980 --> 00:01:34.930
for stakeholders to always look forward to
maximizeing revenue and engagement, for instance.
19
00:01:34.049 --> 00:01:41.489
Why minimizing cost like that product and
customer and right in them. So how
20
00:01:41.609 --> 00:01:45.730
do you first become an above your
industry? So it starts things from my
21
00:01:45.810 --> 00:01:52.560
background, academic and even the educational. So analytics started back with me,
22
00:01:52.680 --> 00:01:57.400
started back from my childhood when I
was already trying to make sense out of
23
00:01:57.480 --> 00:02:02.510
the numbers in the stores. Start
in registers to support my model's poetry firm,
24
00:02:02.989 --> 00:02:08.189
my mother support you firm, and
we looked into registering things in registers
25
00:02:08.310 --> 00:02:13.110
a little bit manually, and I
was already intrigued by that. So the
26
00:02:13.189 --> 00:02:19.699
big data analytics aspect of things came
into play during my capstone project and my
27
00:02:19.860 --> 00:02:25.419
Bachelors Actually, under the title of
this was Predicting College dropouts. So here
28
00:02:25.659 --> 00:02:32.210
this project are actually entailed rundling data
prior to students and these details in huge
29
00:02:32.250 --> 00:02:37.569
amounts, and it literally served as
a point of entry for me in the
30
00:02:37.689 --> 00:02:44.400
big data space, because the project
was to help stakeholders who were the advisors,
31
00:02:44.479 --> 00:02:50.199
academic advisors in this case, to
be able to make insightful decisions and
32
00:02:50.400 --> 00:02:55.479
give good orientations for students in their
courses and, yes, school career,
33
00:02:55.840 --> 00:03:00.870
as to choosing the right faculty to
continue majoring in. So that is how
34
00:03:00.909 --> 00:03:07.430
I started being involved in the big
data space and from here I had a
35
00:03:07.550 --> 00:03:10.189
good number of instructor that we're interested
in this project, for instance, it
36
00:03:10.430 --> 00:03:15.580
ended up being the third place best
Capstone Project for a year twenty seventeen,
37
00:03:15.740 --> 00:03:22.259
and I literally captured the interest of
my instructor at that time, was the
38
00:03:22.620 --> 00:03:27.729
instructs, the instructor of all the
Biggita analytics course, and so I directly
39
00:03:27.849 --> 00:03:34.210
improved my career with the master's degree
in big data analytics, majoring from computer
40
00:03:34.409 --> 00:03:37.650
science. So, that said,
I do have a number of publications in
41
00:03:37.889 --> 00:03:42.120
terms of research and in terms of
articles also. That speaks to big the
42
00:03:42.280 --> 00:03:49.000
analytics ranging from the medical field to
business fields. So my soul John as
43
00:03:49.000 --> 00:03:53.240
a researcher, for instance, and
graduate students, was a resourceful and fruitful
44
00:03:53.319 --> 00:03:59.349
as well, with these research publications
and articles which I did need forward sharing.
45
00:03:59.629 --> 00:04:03.030
Well, yeah, it's always nice
to hear how people became and involved
46
00:04:03.270 --> 00:04:05.789
in data science. or how I
got what I are, and it was
47
00:04:05.830 --> 00:04:10.219
interesting to hear you and about your
mom, and it's predicting college dropouts.
48
00:04:10.460 --> 00:04:14.379
So, regarding your job, what
exactly do you do as a business intelligence
49
00:04:14.419 --> 00:04:18.699
analysts and engineer? Right, so
I'll briefly summarize it, just informing steps
50
00:04:18.779 --> 00:04:24.769
what I do as this intelligence engineer. So first is creating self service tools
51
00:04:24.850 --> 00:04:29.529
for stakeholders right to make insite for
decisions. The second part of it is
52
00:04:29.649 --> 00:04:35.810
identifying and generating key performance indicators for
trend monitoring, right and decisionmaking, giving
53
00:04:35.850 --> 00:04:40.879
that, when there is a lot
of data out there, the the ability
54
00:04:41.040 --> 00:04:45.959
to identify the signal out of noise
is key. The third here is conducting
55
00:04:46.120 --> 00:04:53.350
research and leveraging statistical techniques for customer
segmentation and product positioning, again, as
56
00:04:53.389 --> 00:04:58.790
I previously talk to you about breathing
the gap between products and customers or clients.
57
00:04:59.149 --> 00:05:04.589
And then the fourth is providing descriptive
creditive for prescriptive analytics for informed decision
58
00:05:04.709 --> 00:05:12.420
making. So the added forming qualifiers
or direction givers of what I do was
59
00:05:12.500 --> 00:05:18.500
business intelligence and engineer. It enters
coming up with those right or scripts or,
60
00:05:18.579 --> 00:05:25.290
let's see, stepwise algorithms in order
to carry these from end to end
61
00:05:25.290 --> 00:05:31.490
and provide a solution deployable for relatestickholders. I know data is really probably in
62
00:05:31.569 --> 00:05:38.319
a field and data literacy can be
huge hinderance to building data driven companies.
63
00:05:38.560 --> 00:05:44.439
So how can companies work towards being
data letter it so definitely, especially in
64
00:05:44.560 --> 00:05:46.639
this era and there is a lot
of data, as you rightly said,
65
00:05:46.759 --> 00:05:50.990
there is the need to be able
to channel the right data to be accessed
66
00:05:51.189 --> 00:05:56.870
by the right person and for that
person to make the right thing out of
67
00:05:56.949 --> 00:06:00.910
the data. So I can just
plead this into two main aspects. In
68
00:06:00.110 --> 00:06:04.139
order to enhance data literacy in the
company, the first thing will be to
69
00:06:04.220 --> 00:06:12.180
enable data democratization right true functional governance
structures. So should be able to segment
70
00:06:12.220 --> 00:06:15.300
all the people in the corporate as
to this is what you're doing and this
71
00:06:15.459 --> 00:06:19.730
is the accesses you do have,
for instance under the second part is now
72
00:06:19.889 --> 00:06:26.930
establishing a training platform, for instance, where related stakeholders, are related users
73
00:06:27.089 --> 00:06:30.370
of the of the data can learn
how to use the relevant tools. And
74
00:06:30.810 --> 00:06:35.879
then, regarding a data science problem, what's the step by set step process?
75
00:06:36.000 --> 00:06:39.639
A sumwhere is problem? No matter
what the problem is, is there
76
00:06:39.639 --> 00:06:43.759
a general case, step by step
process. Sure there are a lot.
77
00:06:44.079 --> 00:06:46.120
Prior to my experience when I started, there are a lot of step by
78
00:06:46.189 --> 00:06:51.629
step processes, but with time it's
getting to being compretised and my advice will
79
00:06:51.670 --> 00:06:58.350
be to follow a generally known step
process that will enable you communicate easily with
80
00:06:58.509 --> 00:07:02.100
orders out there in the world when
you want to express an opinion or to
81
00:07:02.220 --> 00:07:05.779
express a step process. For instance, if you use a convention that is
82
00:07:05.819 --> 00:07:12.860
already known, then your way out
there being able to easily understood. So
83
00:07:13.540 --> 00:07:17.569
I am going for the Christ DM
order actually, which is a conventional approach.
84
00:07:17.730 --> 00:07:24.610
Hence it enhances understanding right under the
latter, that is the CRISP DM
85
00:07:24.689 --> 00:07:29.089
model, can be sectioned into six
main steps. The first you have the
86
00:07:29.129 --> 00:07:33.000
business understanding. Yeah, did understanding, your data preparation, your modeling,
87
00:07:33.040 --> 00:07:35.920
that comes into play. You have
evaluation, then you have your resorts.
88
00:07:36.399 --> 00:07:41.519
So following these steps, for instance, you can be able to solve any
89
00:07:41.720 --> 00:07:46.310
form, unkind or structure of the
data science problem. Okay, and it's
90
00:07:46.350 --> 00:07:51.790
called the CRISP DM model. Okay, CHRISTIAN CRISP DM model. Yes,
91
00:07:51.949 --> 00:07:58.980
fine, data science life cycle exactly. Okay, very interesting. So how
92
00:07:59.060 --> 00:08:03.500
do you identify results, choose a
model, and the ploy and evaluate a
93
00:08:03.620 --> 00:08:07.740
model. So this is a good
question. It's a very good one because
94
00:08:07.139 --> 00:08:13.089
it's definitely key questions answer. Whenever
you want to address the business problem,
95
00:08:13.529 --> 00:08:16.930
it gives you direction and also constrained
you on to what to do exactly to
96
00:08:18.089 --> 00:08:22.569
have a problem solved. So in
order to identify results, for instance,
97
00:08:22.889 --> 00:08:30.199
so I will say I identify results
the moment it answers our solutions the business
98
00:08:30.319 --> 00:08:33.759
problem. If something solutions the business
problem, then it is a result for
99
00:08:33.960 --> 00:08:37.639
it. And to choose a model, I choose it by identifying the category
100
00:08:37.679 --> 00:08:41.710
in which the business problem lies.
Again, if you remember the Christian model,
101
00:08:41.750 --> 00:08:45.789
Retord, that what the first thing. The first type is identifying the
102
00:08:45.870 --> 00:08:48.389
business problem. Right. So when
you have a business problem, if you
103
00:08:48.470 --> 00:08:52.070
want to qualify if something is a
result or not the moment, if what
104
00:08:52.269 --> 00:08:56.740
you are trying to qualify answers to
that business problem, then it is a
105
00:08:56.899 --> 00:09:01.220
potential result. And the model is
chosen by categorizing the business problem. Is
106
00:09:01.259 --> 00:09:05.419
it the classification model? Is it
a regression model? Is it an unsupervised
107
00:09:05.460 --> 00:09:11.210
or supervised learning model? I mean
the business problem. How does it categorize
108
00:09:11.409 --> 00:09:15.450
in the various kind of problems we
do have in the data science and analytic
109
00:09:15.570 --> 00:09:18.970
space. That is how you do
choose in model and we can definitely expand
110
00:09:18.049 --> 00:09:22.570
from there as much as possible,
depending on the stickholders knowledge of the the
111
00:09:22.649 --> 00:09:30.159
business problem. Now the models are
deployed based on the tools available to convey
112
00:09:30.240 --> 00:09:35.519
information right under how Literat the stakeholders
are to those tools. If you remember,
113
00:09:37.039 --> 00:09:41.429
the great questioning you you ask prior
was how to enhance the Dai tracing
114
00:09:41.509 --> 00:09:45.990
the good in the company. Now, depending on how literate company is,
115
00:09:46.149 --> 00:09:48.309
you can see what kind of tools
you can leave reage in order to append
116
00:09:48.350 --> 00:09:52.580
to such a business problem and in
order to deploy your model. It can
117
00:09:52.700 --> 00:09:58.179
range from dataware houses to data leaks
or to a simple data base, depending
118
00:09:58.220 --> 00:10:03.019
on the structure of your company.
Now to evaluate your model. They are
119
00:10:03.059 --> 00:10:09.169
various evaluation techniques hour there in the
world and the evolution technique used generally depends
120
00:10:09.330 --> 00:10:16.090
is on the model type and whether
it provides an optimal solution to the business
121
00:10:16.250 --> 00:10:20.009
problem or not. I will take
a simple example. So you can have
122
00:10:20.009 --> 00:10:24.960
model type which is of classification type. That means the business model is looking
123
00:10:26.039 --> 00:10:30.279
forward to class of fine, of
classic listing, Binary classificient zero or one,
124
00:10:30.519 --> 00:10:33.200
and so you can go you are. You would already be constrained into
125
00:10:33.279 --> 00:10:39.350
a set of evaluation techniques. Right. Could be moving from accuracy precision,
126
00:10:39.590 --> 00:10:45.389
recall or maybe f one score and
now to have an optimal solution for your
127
00:10:45.470 --> 00:10:50.860
business problem, you could either decide
to focus on recall or precision. So
128
00:10:50.259 --> 00:10:56.100
that is how I look forward to
evaluating a model. Does it your questions?
129
00:10:56.860 --> 00:10:58.220
That was good. Now, yes, thank you for your insight.
130
00:10:58.379 --> 00:11:01.100
I just listening way. Have to
say I do want to go back to
131
00:11:01.220 --> 00:11:05.250
the CRISP DM. Can please explain
that a little bit further? Oh yeah,
132
00:11:05.610 --> 00:11:11.730
the Christium model stands for cross industry
standard process, but a mining so,
133
00:11:13.169 --> 00:11:16.490
as the name sees, as an
acronym says, is this the standard
134
00:11:16.529 --> 00:11:20.240
process that is outer in the world
already for you to kind of follow in
135
00:11:20.360 --> 00:11:24.519
order to help or the data scientist
easily conveyed with you under make sure they
136
00:11:24.639 --> 00:11:30.240
just tag at the end right and
not look as to how do you do
137
00:11:30.480 --> 00:11:33.509
this? How do you do that? Is just a structured out there in
138
00:11:33.549 --> 00:11:39.070
the world that helps enable understanding.
So that's said. Given the six steps
139
00:11:39.149 --> 00:11:43.909
we just described, the Christ DM
model is a great future and it's of
140
00:11:43.029 --> 00:11:50.500
great use to promote the common language
for data scientists and enhance communication, reproducibility
141
00:11:50.059 --> 00:11:54.620
and the efficiency. I would say, for instance, the Christian model is
142
00:11:54.700 --> 00:12:00.539
similar to is to issue the data
science program solving cycle, just as the
143
00:12:00.899 --> 00:12:05.090
software deloment cycle is to for software
developers. If there's a good analogy,
144
00:12:05.769 --> 00:12:11.330
okay, cant that one. Okay. Well, is there a greatest challenge
145
00:12:11.370 --> 00:12:16.519
and your industry that you see a
great challenge in the industry that I see?
146
00:12:16.120 --> 00:12:18.799
I would see. The first one
I had, which I saw,
147
00:12:18.919 --> 00:12:24.679
which is getting solved, was the
the the big data boom. Like those
148
00:12:24.679 --> 00:12:28.840
a lot of data and those need
to come up with resources to be able
149
00:12:28.879 --> 00:12:31.669
to process that data. But now
we are an in an era where the
150
00:12:31.830 --> 00:12:37.309
resources are really available to process this
kind, this huge amount of data.
151
00:12:37.789 --> 00:12:41.350
We do have hood, for instance
in media, that comes in with the
152
00:12:41.509 --> 00:12:46.220
powerl processors to be able to parallel, to to to process and parallel huge
153
00:12:46.259 --> 00:12:50.460
chance of data. We do our
systems such as datags from the DA warehouse,
154
00:12:50.539 --> 00:12:52.940
is to be dialect to be able
to analyze data in real time.
155
00:12:54.460 --> 00:12:58.500
That said, they are features.
Are The already that helps you address the
156
00:12:58.730 --> 00:13:03.610
velocity variety aspect of the data.
So the main issue. Now for me
157
00:13:03.769 --> 00:13:07.570
I think it's the being able to
curate data and identify the signal out of
158
00:13:07.730 --> 00:13:13.679
the noise. That is the main
challenge in the industry. And One pitfall
159
00:13:13.840 --> 00:13:16.879
is that a lot of industries,
or let's say a lot of people,
160
00:13:16.080 --> 00:13:20.799
think moment you have big data,
then you have a solution to a program
161
00:13:20.840 --> 00:13:26.360
which is not necessarily true. You
can also use a small data set that
162
00:13:26.519 --> 00:13:30.549
the moment it is efficient and of
well cleaned up and curated, you can
163
00:13:30.590 --> 00:13:33.429
still attain to the same result.
So I would say that as a challenge
164
00:13:33.629 --> 00:13:39.590
to identify the signal amount of the
noise ending it was a challenge. I
165
00:13:39.789 --> 00:13:43.779
like it. So for more information
on a data standard, you can find
166
00:13:43.820 --> 00:13:50.340
us at taww. That data standard
dial also, this episode was sponsored by
167
00:13:50.460 --> 00:13:54.059
Pandel and they're innovating the text base
and you can learn more about their work
168
00:13:54.139 --> 00:13:58.169
at pandiocom. But NOWSON, they
give for joining us and where can we
169
00:13:58.289 --> 00:14:01.330
find you? Are you on Linkedin? Sure, I'm on Linkedin. My
170
00:14:01.450 --> 00:14:03.769
linkedin is just my first in mind. My last names up Nelson. At
171
00:14:03.850 --> 00:14:07.090
this stage, I would see giving
back to the community is a trend,
172
00:14:07.250 --> 00:14:13.519
is of treaming those importance not only
to grow relivant networks right, but also
173
00:14:13.600 --> 00:14:16.759
to master the concept and to master
the field. So I do have a
174
00:14:16.799 --> 00:14:22.840
good number of relivant research publications,
not articles which I wound mind sharing,
175
00:14:22.000 --> 00:14:26.429
you know, and the yeah,
hence I mean I'll be a big glad
176
00:14:26.509 --> 00:14:31.070
seeing the community reaping the foods out
of this. So I love a linkedin.
177
00:14:31.110 --> 00:14:33.350
I also go tweeter accounts. I
will definitely miss any back to you
178
00:14:33.509 --> 00:14:39.059
the already after interview. Any of
publications? Yeah, I do have two
179
00:14:39.100 --> 00:14:46.259
major publications which you will definitely find
interesting. One is the recommendation engines,
180
00:14:46.460 --> 00:14:52.740
boosting recommendation engines using an offline evaluation
metric approach. So this, people,
181
00:14:52.779 --> 00:14:58.649
was actually published at a CMC ACM
in two thousand eighteen and it's out there
182
00:15:00.370 --> 00:15:05.490
already. And another one is also
on about informatics. It's actually a machine
183
00:15:05.490 --> 00:15:13.000
learning texture based approached to be able
to detect cancer from too much epitaial source.
184
00:15:13.600 --> 00:15:15.840
So, if you remember the beginning
of the talk, I talk to
185
00:15:15.879 --> 00:15:22.029
you moving from the bowl informatic space
right to business space. So the second
186
00:15:22.070 --> 00:15:24.629
pepius mentioned is much more in VII
and formatics, which is biology, and
187
00:15:24.750 --> 00:15:31.870
the first, the first, which
is boosting recommendations speaks to the business understanding
188
00:15:31.029 --> 00:15:35.860
face. So now, for instance, using recommendation engines. You can think
189
00:15:35.940 --> 00:15:39.659
of Netflix, or instance, and
the funk as with the model that is
190
00:15:39.700 --> 00:15:43.899
being used to recommend movies. You
can think of Amazon, for instance,
191
00:15:45.419 --> 00:15:50.210
the technologies and the tools they use
to recommend other products for people who bought
192
00:15:50.330 --> 00:15:58.009
some other products. Also recommending things
to people who had friends doing cross and
193
00:15:58.090 --> 00:16:03.570
by productor commendations just a part of
it. Wow, this is so great.
194
00:16:03.690 --> 00:16:07.000
I don't work and we find your
publications. If you get my linked
195
00:16:07.039 --> 00:16:10.799
in, which I would definitely share
with you, in the introduction you have
196
00:16:10.960 --> 00:16:15.440
the name of the two papers and
you have the publications and you have the
197
00:16:15.480 --> 00:16:19.789
dates and also in the publications section
of Linkedin you can also see the two
198
00:16:19.909 --> 00:16:25.389
links of the two main publications.
Then there are five of them overall,
199
00:16:25.710 --> 00:16:29.909
but those fer to give a brief
summary. Okay, well, sounds great
200
00:16:30.070 --> 00:16:33.100
and thank you so much for taking
the time to speak of there. Sure
201
00:16:33.259 --> 00:16:34.940
it was my pleasure. China,
YEA, and thanks to show your insight.
202
00:16:36.019 --> 00:16:37.740
A lot prospecting of here. So
thank you, no problem. You
203
00:16:37.820 --> 00:16:38.500
have a good one.
1
00:00:10.310 --> 00:00:14.349
Hello data standard community. I'm your
host, Shane Waldon, and this is
2
00:00:14.390 --> 00:00:18.149
the data standard audio experience. We
have a very exciting guest today on our
3
00:00:18.190 --> 00:00:23.100
podcast. We have Nelson Saffo,
a digital business intelligence analyst, but the
4
00:00:23.219 --> 00:00:28.179
Atlanta Journal Constitution at Cok Center.
Prise Cock drives new ways of innovation in
5
00:00:28.260 --> 00:00:33.340
the communication and animated industries, and
Nelson has specialties and machine learning, big
6
00:00:33.380 --> 00:00:37.810
day analytics and etl pipelines. Welcome
to the show, Nelson. Thank you
7
00:00:38.009 --> 00:00:43.929
for the good introduction machine. It's
a pleasure talking with you at this point
8
00:00:43.969 --> 00:00:47.570
in time. Thank you. I
am so excited to get another better and
9
00:00:47.689 --> 00:00:50.840
I start off. Can you please
just introduce yourself? Sure, it's a
10
00:00:50.880 --> 00:00:54.719
pleasure talking with you at this point
in time. Shan Yea and the first
11
00:00:54.759 --> 00:00:58.560
I just wanted to clarify. I'm
no longer we talk. has been a
12
00:00:58.679 --> 00:01:02.320
month already, as I look forward
to an advancement in Korea, but of
13
00:01:02.439 --> 00:01:07.189
course, my experience in the mid
of cogs is very happy to share with
14
00:01:07.310 --> 00:01:11.109
you what I've been doing, especially
as this podcast is prior to eat.
15
00:01:11.790 --> 00:01:15.670
So I will just give a brief
some areas an introduction, as the one
16
00:01:15.750 --> 00:01:21.219
sentence description of me, who is
this data study. Individual with takes pleasure
17
00:01:21.379 --> 00:01:26.780
in reaching the gap between customers and
products, with insight for this aision making
18
00:01:26.980 --> 00:01:34.930
for stakeholders to always look forward to
maximizeing revenue and engagement, for instance.
19
00:01:34.049 --> 00:01:41.489
Why minimizing cost like that product and
customer and right in them. So how
20
00:01:41.609 --> 00:01:45.730
do you first become an above your
industry? So it starts things from my
21
00:01:45.810 --> 00:01:52.560
background, academic and even the educational. So analytics started back with me,
22
00:01:52.680 --> 00:01:57.400
started back from my childhood when I
was already trying to make sense out of
23
00:01:57.480 --> 00:02:02.510
the numbers in the stores. Start
in registers to support my model's poetry firm,
24
00:02:02.989 --> 00:02:08.189
my mother support you firm, and
we looked into registering things in registers
25
00:02:08.310 --> 00:02:13.110
a little bit manually, and I
was already intrigued by that. So the
26
00:02:13.189 --> 00:02:19.699
big data analytics aspect of things came
into play during my capstone project and my
27
00:02:19.860 --> 00:02:25.419
Bachelors Actually, under the title of
this was Predicting College dropouts. So here
28
00:02:25.659 --> 00:02:32.210
this project are actually entailed rundling data
prior to students and these details in huge
29
00:02:32.250 --> 00:02:37.569
amounts, and it literally served as
a point of entry for me in the
30
00:02:37.689 --> 00:02:44.400
big data space, because the project
was to help stakeholders who were the advisors,
31
00:02:44.479 --> 00:02:50.199
academic advisors in this case, to
be able to make insightful decisions and
32
00:02:50.400 --> 00:02:55.479
give good orientations for students in their
courses and, yes, school career,
33
00:02:55.840 --> 00:03:00.870
as to choosing the right faculty to
continue majoring in. So that is how
34
00:03:00.909 --> 00:03:07.430
I started being involved in the big
data space and from here I had a
35
00:03:07.550 --> 00:03:10.189
good number of instructor that we're interested
in this project, for instance, it
36
00:03:10.430 --> 00:03:15.580
ended up being the third place best
Capstone Project for a year twenty seventeen,
37
00:03:15.740 --> 00:03:22.259
and I literally captured the interest of
my instructor at that time, was the
38
00:03:22.620 --> 00:03:27.729
instructs, the instructor of all the
Biggita analytics course, and so I directly
39
00:03:27.849 --> 00:03:34.210
improved my career with the master's degree
in big data analytics, majoring from computer
40
00:03:34.409 --> 00:03:37.650
science. So, that said,
I do have a number of publications in
41
00:03:37.889 --> 00:03:42.120
terms of research and in terms of
articles also. That speaks to big the
42
00:03:42.280 --> 00:03:49.000
analytics ranging from the medical field to
business fields. So my soul John as
43
00:03:49.000 --> 00:03:53.240
a researcher, for instance, and
graduate students, was a resourceful and fruitful
44
00:03:53.319 --> 00:03:59.349
as well, with these research publications
and articles which I did need forward sharing.
45
00:03:59.629 --> 00:04:03.030
Well, yeah, it's always nice
to hear how people became and involved
46
00:04:03.270 --> 00:04:05.789
in data science. or how I
got what I are, and it was
47
00:04:05.830 --> 00:04:10.219
interesting to hear you and about your
mom, and it's predicting college dropouts.
48
00:04:10.460 --> 00:04:14.379
So, regarding your job, what
exactly do you do as a business intelligence
49
00:04:14.419 --> 00:04:18.699
analysts and engineer? Right, so
I'll briefly summarize it, just informing steps
50
00:04:18.779 --> 00:04:24.769
what I do as this intelligence engineer. So first is creating self service tools
51
00:04:24.850 --> 00:04:29.529
for stakeholders right to make insite for
decisions. The second part of it is
52
00:04:29.649 --> 00:04:35.810
identifying and generating key performance indicators for
trend monitoring, right and decisionmaking, giving
53
00:04:35.850 --> 00:04:40.879
that, when there is a lot
of data out there, the the ability
54
00:04:41.040 --> 00:04:45.959
to identify the signal out of noise
is key. The third here is conducting
55
00:04:46.120 --> 00:04:53.350
research and leveraging statistical techniques for customer
segmentation and product positioning, again, as
56
00:04:53.389 --> 00:04:58.790
I previously talk to you about breathing
the gap between products and customers or clients.
57
00:04:59.149 --> 00:05:04.589
And then the fourth is providing descriptive
creditive for prescriptive analytics for informed decision
58
00:05:04.709 --> 00:05:12.420
making. So the added forming qualifiers
or direction givers of what I do was
59
00:05:12.500 --> 00:05:18.500
business intelligence and engineer. It enters
coming up with those right or scripts or,
60
00:05:18.579 --> 00:05:25.290
let's see, stepwise algorithms in order
to carry these from end to end
61
00:05:25.290 --> 00:05:31.490
and provide a solution deployable for relatestickholders. I know data is really probably in
62
00:05:31.569 --> 00:05:38.319
a field and data literacy can be
huge hinderance to building data driven companies.
63
00:05:38.560 --> 00:05:44.439
So how can companies work towards being
data letter it so definitely, especially in
64
00:05:44.560 --> 00:05:46.639
this era and there is a lot
of data, as you rightly said,
65
00:05:46.759 --> 00:05:50.990
there is the need to be able
to channel the right data to be accessed
66
00:05:51.189 --> 00:05:56.870
by the right person and for that
person to make the right thing out of
67
00:05:56.949 --> 00:06:00.910
the data. So I can just
plead this into two main aspects. In
68
00:06:00.110 --> 00:06:04.139
order to enhance data literacy in the
company, the first thing will be to
69
00:06:04.220 --> 00:06:12.180
enable data democratization right true functional governance
structures. So should be able to segment
70
00:06:12.220 --> 00:06:15.300
all the people in the corporate as
to this is what you're doing and this
71
00:06:15.459 --> 00:06:19.730
is the accesses you do have,
for instance under the second part is now
72
00:06:19.889 --> 00:06:26.930
establishing a training platform, for instance, where related stakeholders, are related users
73
00:06:27.089 --> 00:06:30.370
of the of the data can learn
how to use the relevant tools. And
74
00:06:30.810 --> 00:06:35.879
then, regarding a data science problem, what's the step by set step process?
75
00:06:36.000 --> 00:06:39.639
A sumwhere is problem? No matter
what the problem is, is there
76
00:06:39.639 --> 00:06:43.759
a general case, step by step
process. Sure there are a lot.
77
00:06:44.079 --> 00:06:46.120
Prior to my experience when I started, there are a lot of step by
78
00:06:46.189 --> 00:06:51.629
step processes, but with time it's
getting to being compretised and my advice will
79
00:06:51.670 --> 00:06:58.350
be to follow a generally known step
process that will enable you communicate easily with
80
00:06:58.509 --> 00:07:02.100
orders out there in the world when
you want to express an opinion or to
81
00:07:02.220 --> 00:07:05.779
express a step process. For instance, if you use a convention that is
82
00:07:05.819 --> 00:07:12.860
already known, then your way out
there being able to easily understood. So
83
00:07:13.540 --> 00:07:17.569
I am going for the Christ DM
order actually, which is a conventional approach.
84
00:07:17.730 --> 00:07:24.610
Hence it enhances understanding right under the
latter, that is the CRISP DM
85
00:07:24.689 --> 00:07:29.089
model, can be sectioned into six
main steps. The first you have the
86
00:07:29.129 --> 00:07:33.000
business understanding. Yeah, did understanding, your data preparation, your modeling,
87
00:07:33.040 --> 00:07:35.920
that comes into play. You have
evaluation, then you have your resorts.
88
00:07:36.399 --> 00:07:41.519
So following these steps, for instance, you can be able to solve any
89
00:07:41.720 --> 00:07:46.310
form, unkind or structure of the
data science problem. Okay, and it's
90
00:07:46.350 --> 00:07:51.790
called the CRISP DM model. Okay, CHRISTIAN CRISP DM model. Yes,
91
00:07:51.949 --> 00:07:58.980
fine, data science life cycle exactly. Okay, very interesting. So how
92
00:07:59.060 --> 00:08:03.500
do you identify results, choose a
model, and the ploy and evaluate a
93
00:08:03.620 --> 00:08:07.740
model. So this is a good
question. It's a very good one because
94
00:08:07.139 --> 00:08:13.089
it's definitely key questions answer. Whenever
you want to address the business problem,
95
00:08:13.529 --> 00:08:16.930
it gives you direction and also constrained
you on to what to do exactly to
96
00:08:18.089 --> 00:08:22.569
have a problem solved. So in
order to identify results, for instance,
97
00:08:22.889 --> 00:08:30.199
so I will say I identify results
the moment it answers our solutions the business
98
00:08:30.319 --> 00:08:33.759
problem. If something solutions the business
problem, then it is a result for
99
00:08:33.960 --> 00:08:37.639
it. And to choose a model, I choose it by identifying the category
100
00:08:37.679 --> 00:08:41.710
in which the business problem lies.
Again, if you remember the Christian model,
101
00:08:41.750 --> 00:08:45.789
Retord, that what the first thing. The first type is identifying the
102
00:08:45.870 --> 00:08:48.389
business problem. Right. So when
you have a business problem, if you
103
00:08:48.470 --> 00:08:52.070
want to qualify if something is a
result or not the moment, if what
104
00:08:52.269 --> 00:08:56.740
you are trying to qualify answers to
that business problem, then it is a
105
00:08:56.899 --> 00:09:01.220
potential result. And the model is
chosen by categorizing the business problem. Is
106
00:09:01.259 --> 00:09:05.419
it the classification model? Is it
a regression model? Is it an unsupervised
107
00:09:05.460 --> 00:09:11.210
or supervised learning model? I mean
the business problem. How does it categorize
108
00:09:11.409 --> 00:09:15.450
in the various kind of problems we
do have in the data science and analytic
109
00:09:15.570 --> 00:09:18.970
space. That is how you do
choose in model and we can definitely expand
110
00:09:18.049 --> 00:09:22.570
from there as much as possible,
depending on the stickholders knowledge of the the
111
00:09:22.649 --> 00:09:30.159
business problem. Now the models are
deployed based on the tools available to convey
112
00:09:30.240 --> 00:09:35.519
information right under how Literat the stakeholders
are to those tools. If you remember,
113
00:09:37.039 --> 00:09:41.429
the great questioning you you ask prior
was how to enhance the Dai tracing
114
00:09:41.509 --> 00:09:45.990
the good in the company. Now, depending on how literate company is,
115
00:09:46.149 --> 00:09:48.309
you can see what kind of tools
you can leave reage in order to append
116
00:09:48.350 --> 00:09:52.580
to such a business problem and in
order to deploy your model. It can
117
00:09:52.700 --> 00:09:58.179
range from dataware houses to data leaks
or to a simple data base, depending
118
00:09:58.220 --> 00:10:03.019
on the structure of your company.
Now to evaluate your model. They are
119
00:10:03.059 --> 00:10:09.169
various evaluation techniques hour there in the
world and the evolution technique used generally depends
120
00:10:09.330 --> 00:10:16.090
is on the model type and whether
it provides an optimal solution to the business
121
00:10:16.250 --> 00:10:20.009
problem or not. I will take
a simple example. So you can have
122
00:10:20.009 --> 00:10:24.960
model type which is of classification type. That means the business model is looking
123
00:10:26.039 --> 00:10:30.279
forward to class of fine, of
classic listing, Binary classificient zero or one,
124
00:10:30.519 --> 00:10:33.200
and so you can go you are. You would already be constrained into
125
00:10:33.279 --> 00:10:39.350
a set of evaluation techniques. Right. Could be moving from accuracy precision,
126
00:10:39.590 --> 00:10:45.389
recall or maybe f one score and
now to have an optimal solution for your
127
00:10:45.470 --> 00:10:50.860
business problem, you could either decide
to focus on recall or precision. So
128
00:10:50.259 --> 00:10:56.100
that is how I look forward to
evaluating a model. Does it your questions?
129
00:10:56.860 --> 00:10:58.220
That was good. Now, yes, thank you for your insight.
130
00:10:58.379 --> 00:11:01.100
I just listening way. Have to
say I do want to go back to
131
00:11:01.220 --> 00:11:05.250
the CRISP DM. Can please explain
that a little bit further? Oh yeah,
132
00:11:05.610 --> 00:11:11.730
the Christium model stands for cross industry
standard process, but a mining so,
133
00:11:13.169 --> 00:11:16.490
as the name sees, as an
acronym says, is this the standard
134
00:11:16.529 --> 00:11:20.240
process that is outer in the world
already for you to kind of follow in
135
00:11:20.360 --> 00:11:24.519
order to help or the data scientist
easily conveyed with you under make sure they
136
00:11:24.639 --> 00:11:30.240
just tag at the end right and
not look as to how do you do
137
00:11:30.480 --> 00:11:33.509
this? How do you do that? Is just a structured out there in
138
00:11:33.549 --> 00:11:39.070
the world that helps enable understanding.
So that's said. Given the six steps
139
00:11:39.149 --> 00:11:43.909
we just described, the Christ DM
model is a great future and it's of
140
00:11:43.029 --> 00:11:50.500
great use to promote the common language
for data scientists and enhance communication, reproducibility
141
00:11:50.059 --> 00:11:54.620
and the efficiency. I would say, for instance, the Christian model is
142
00:11:54.700 --> 00:12:00.539
similar to is to issue the data
science program solving cycle, just as the
143
00:12:00.899 --> 00:12:05.090
software deloment cycle is to for software
developers. If there's a good analogy,
144
00:12:05.769 --> 00:12:11.330
okay, cant that one. Okay. Well, is there a greatest challenge
145
00:12:11.370 --> 00:12:16.519
and your industry that you see a
great challenge in the industry that I see?
146
00:12:16.120 --> 00:12:18.799
I would see. The first one
I had, which I saw,
147
00:12:18.919 --> 00:12:24.679
which is getting solved, was the
the the big data boom. Like those
148
00:12:24.679 --> 00:12:28.840
a lot of data and those need
to come up with resources to be able
149
00:12:28.879 --> 00:12:31.669
to process that data. But now
we are an in an era where the
150
00:12:31.830 --> 00:12:37.309
resources are really available to process this
kind, this huge amount of data.
151
00:12:37.789 --> 00:12:41.350
We do have hood, for instance
in media, that comes in with the
152
00:12:41.509 --> 00:12:46.220
powerl processors to be able to parallel, to to to process and parallel huge
153
00:12:46.259 --> 00:12:50.460
chance of data. We do our
systems such as datags from the DA warehouse,
154
00:12:50.539 --> 00:12:52.940
is to be dialect to be able
to analyze data in real time.
155
00:12:54.460 --> 00:12:58.500
That said, they are features.
Are The already that helps you address the
156
00:12:58.730 --> 00:13:03.610
velocity variety aspect of the data.
So the main issue. Now for me
157
00:13:03.769 --> 00:13:07.570
I think it's the being able to
curate data and identify the signal out of
158
00:13:07.730 --> 00:13:13.679
the noise. That is the main
challenge in the industry. And One pitfall
159
00:13:13.840 --> 00:13:16.879
is that a lot of industries,
or let's say a lot of people,
160
00:13:16.080 --> 00:13:20.799
think moment you have big data,
then you have a solution to a program
161
00:13:20.840 --> 00:13:26.360
which is not necessarily true. You
can also use a small data set that
162
00:13:26.519 --> 00:13:30.549
the moment it is efficient and of
well cleaned up and curated, you can
163
00:13:30.590 --> 00:13:33.429
still attain to the same result.
So I would say that as a challenge
164
00:13:33.629 --> 00:13:39.590
to identify the signal amount of the
noise ending it was a challenge. I
165
00:13:39.789 --> 00:13:43.779
like it. So for more information
on a data standard, you can find
166
00:13:43.820 --> 00:13:50.340
us at taww. That data standard
dial also, this episode was sponsored by
167
00:13:50.460 --> 00:13:54.059
Pandel and they're innovating the text base
and you can learn more about their work
168
00:13:54.139 --> 00:13:58.169
at pandiocom. But NOWSON, they
give for joining us and where can we
169
00:13:58.289 --> 00:14:01.330
find you? Are you on Linkedin? Sure, I'm on Linkedin. My
170
00:14:01.450 --> 00:14:03.769
linkedin is just my first in mind. My last names up Nelson. At
171
00:14:03.850 --> 00:14:07.090
this stage, I would see giving
back to the community is a trend,
172
00:14:07.250 --> 00:14:13.519
is of treaming those importance not only
to grow relivant networks right, but also
173
00:14:13.600 --> 00:14:16.759
to master the concept and to master
the field. So I do have a
174
00:14:16.799 --> 00:14:22.840
good number of relivant research publications,
not articles which I wound mind sharing,
175
00:14:22.000 --> 00:14:26.429
you know, and the yeah,
hence I mean I'll be a big glad
176
00:14:26.509 --> 00:14:31.070
seeing the community reaping the foods out
of this. So I love a linkedin.
177
00:14:31.110 --> 00:14:33.350
I also go tweeter accounts. I
will definitely miss any back to you
178
00:14:33.509 --> 00:14:39.059
the already after interview. Any of
publications? Yeah, I do have two
179
00:14:39.100 --> 00:14:46.259
major publications which you will definitely find
interesting. One is the recommendation engines,
180
00:14:46.460 --> 00:14:52.740
boosting recommendation engines using an offline evaluation
metric approach. So this, people,
181
00:14:52.779 --> 00:14:58.649
was actually published at a CMC ACM
in two thousand eighteen and it's out there
182
00:15:00.370 --> 00:15:05.490
already. And another one is also
on about informatics. It's actually a machine
183
00:15:05.490 --> 00:15:13.000
learning texture based approached to be able
to detect cancer from too much epitaial source.
184
00:15:13.600 --> 00:15:15.840
So, if you remember the beginning
of the talk, I talk to
185
00:15:15.879 --> 00:15:22.029
you moving from the bowl informatic space
right to business space. So the second
186
00:15:22.070 --> 00:15:24.629
pepius mentioned is much more in VII
and formatics, which is biology, and
187
00:15:24.750 --> 00:15:31.870
the first, the first, which
is boosting recommendations speaks to the business understanding
188
00:15:31.029 --> 00:15:35.860
face. So now, for instance, using recommendation engines. You can think
189
00:15:35.940 --> 00:15:39.659
of Netflix, or instance, and
the funk as with the model that is
190
00:15:39.700 --> 00:15:43.899
being used to recommend movies. You
can think of Amazon, for instance,
191
00:15:45.419 --> 00:15:50.210
the technologies and the tools they use
to recommend other products for people who bought
192
00:15:50.330 --> 00:15:58.009
some other products. Also recommending things
to people who had friends doing cross and
193
00:15:58.090 --> 00:16:03.570
by productor commendations just a part of
it. Wow, this is so great.
194
00:16:03.690 --> 00:16:07.000
I don't work and we find your
publications. If you get my linked
195
00:16:07.039 --> 00:16:10.799
in, which I would definitely share
with you, in the introduction you have
196
00:16:10.960 --> 00:16:15.440
the name of the two papers and
you have the publications and you have the
197
00:16:15.480 --> 00:16:19.789
dates and also in the publications section
of Linkedin you can also see the two
198
00:16:19.909 --> 00:16:25.389
links of the two main publications.
Then there are five of them overall,
199
00:16:25.710 --> 00:16:29.909
but those fer to give a brief
summary. Okay, well, sounds great
200
00:16:30.070 --> 00:16:33.100
and thank you so much for taking
the time to speak of there. Sure
201
00:16:33.259 --> 00:16:34.940
it was my pleasure. China,
YEA, and thanks to show your insight.
202
00:16:36.019 --> 00:16:37.740
A lot prospecting of here. So
thank you, no problem. You
203
00:16:37.820 --> 00:16:38.500
have a good one.
