Data Science: From Raw Data to Decisions

Useful analysis is not the most sophisticated, it is the kind that changes a decision. Here is how to do it properly.

From work we have handled

We mapped the Gorontalo provincial economy by gathering data from Google Maps and several major marketplaces, alongside analysis work for research and business needs. Including unwelcome findings, such as zero Gorontalo-based sellers across 750 products checked on one marketplace.

What makes analysis useful

Data science is not about producing attractive charts, it is about answering questions whose answers change what you do. If you would act identically after seeing the result, the analysis was useless however neat it looked.

So the work starts from a question, not from data. "Which products should we stop stocking" is useful. "Have a look at the sales data" is not, and analysis that begins there usually ends as a dashboard nobody opens again.

When analysis will not help

If data was never recorded consistently, no technique repairs it, results will merely look convincing while remaining wrong. If the decision is already made and analysis is being sought to justify it, that is not analysis. With very little data, any conclusion is easily an accident. And if you need to know why something happened, historical data is often insufficient, that requires a designed experiment.

The workflow

The time sink is not the analysis, it is cleaning. A common estimate puts roughly 80% of effort there.

1Write down the question and what you would do for each possible answer. If the action is the same either way, stop here.
2Trace where the data comes from. Who enters it, when, and what happens when it is blank, these questions expose big problems early.
3Clean it, and record every cleaning decision. Rows dropped and values filled must be defensible later.
4Look at the distribution before computing averages. A single extreme value can make an average entirely misleading.
5Check whether the pattern holds across different slices. A pattern appearing in only one period is usually coincidence.
6Present in the language of decisions, not statistics. Decision-makers need a recommended action and its risk.

Traps that produce wrong conclusions

These four often slip through review and end up as bad decisions:

Treating correlation as cause. Ice cream sales and drowning incidents rise together; neither causes the other.
Assuming missing data is random. Customers who skip the survey are often the least satisfied, ignoring them flatters the result.
Digging until a desired pattern appears. Test enough things and something will always look meaningful by chance.
Looking only at what remains. Analysing customers who stayed without examining those who left misses the reason they left.

Was this article helpful?

Available courses

Kursus Python: Machine Learning Data Science Menggunakan Python & R

Kelas Machine Learning Data Science Menggunakan Python & R untuk analis data, mahasiswa riset, dan tim bisnis yang ingin mengubah data mentah jadi insight yang bisa dipakai ambil keputusan. Kursus Python ini tersedia juga sebagai training Python maupun pelatihan Python in-house untuk kebutuhan perusahaan.

Kursus Data Science: Implementasi Deep Learning untuk Deteksi Emosi dari Teks Menggunakan BERT dan Python

Data Science untuk mengubah data mentah jadi insight yang bisa dipakai ambil keputusan — fokus di Implementasi Deep Learning untuk Deteksi Emosi dari Teks Menggunakan BERT dan Python. Kursus live via Zoom untuk analis data, mahasiswa riset, dan tim bisnis. Kursus Data Science ini tersedia juga sebagai training Data Science maupun pelatihan Data Science in-house untuk kebutuhan perusahaan.

Kursus Python: Data Science 2021: Data Science & Machine Learning Python

Data Science 2021: Data Science & Machine Learning Python untuk mengubah data mentah jadi insight yang bisa dipakai ambil keputusan. Kursus live via Zoom untuk analis data, mahasiswa riset, dan tim bisnis. Kursus Python ini tersedia juga sebagai training Python maupun pelatihan Python in-house untuk kebutuhan perusahaan.

Kursus Data Science: Bootcamp Pembelajaran Mesin Otomatis (ML Otomatis) Membangun 15 Proyek

Data Science untuk mengubah data mentah jadi insight yang bisa dipakai ambil keputusan — fokus di Bootcamp Pembelajaran Mesin Otomatis (ML Otomatis) Membangun 15 Proyek. Kursus live via Zoom untuk analis data, mahasiswa riset, dan tim bisnis. Kursus Data Science ini tersedia juga sebagai training Data Science maupun pelatihan Data Science in-house untuk kebutuhan perusahaan.

Kursus SAS: Complete & Practical SAS, Statistics, Data Science & Data Analysis Course

Kelas Complete & Practical SAS, Statistics, Data Science & Data Analysis Course untuk analis data, mahasiswa riset, dan tim bisnis yang ingin mengubah data mentah jadi insight yang bisa dipakai ambil keputusan. Kursus SAS ini tersedia juga sebagai training SAS maupun pelatihan SAS in-house untuk kebutuhan perusahaan.

Kursus Data Science: Ilmu Data & Pembelajaran Mesin: Hands on Data Science 2020

Belajar Ilmu Data & Pembelajaran Mesin: Hands on Data Science 2020 lewat praktik langsung, dibimbing instruktur via Zoom. Disusun untuk analis data, mahasiswa riset, dan tim bisnis. Kursus Data Science ini tersedia juga sebagai training Data Science maupun pelatihan Data Science in-house untuk kebutuhan perusahaan.

See all related courses

Prefer not to build it yourself?

We also take on projects, from planning through to a working system.

Jasa Pembuatan Python — Data Science & Deep Learning Untuk Business 20 Case Studies

Jasa Pembuatan Data Science ~ Analisis Sentimen Multibahasa pada Twitter

Jasa Pembuatan Data Science_Machine Learning Dengan Pemrograman R

Jasa Pembuatan Python — Data Science, Machine Learning Dan Data Analysis Menggunakan Python

Comments (0)

No comments yet. Be the first to comment.

Talk through your needs first

Tell us what you need and the context it will be used in. We will tell you whether this approach genuinely fits your case -- including if we think there is a simpler or cheaper way.

Consult via WhatsApp

Memuat