Python ile Otomatik Veri Analizi

Turning analysis into a workflow

Automated data analysis makes the repeating steps — loading, quality checks, summaries and reporting — defined and re-runnable. On daily sales, survey or usage data it saves serious time. The one thing it does not save is interpretation: reading the context and investigating surprising findings is still the analyst's job.

1. Load the data and look at it

Declare the encoding and the date columns explicitly. Review the shape, a few sample rows and the dtypes at the very first step.

from pathlib import Path
import pandas as pd

dosya = Path("veri/siparisler.csv")
df = pd.read_csv(dosya, parse_dates=["siparis_tarihi"], encoding="utf-8")
print(df.shape, df.head(), df.dtypes)

beklenen = {"siparis_tarihi", "bolge", "adet", "birim_fiyat"}
eksik = beklenen - set(df.columns)
if eksik:
    raise ValueError(f"Missing expected columns: {sorted(eksik)}")

That schema check is small but critical: when the source file changes quietly, it stops the script on the first line instead of letting you discover weeks later that a report was built on missing fields.

2. Check data quality

Measure missing fields, duplicate records, column types and values that break the business rules. This step is not the analysis itself — it is the answer to whether you can trust the analysis.

profil = pd.DataFrame({
    "tur": df.dtypes.astype("string"),
    "eksik_sayi": df.isna().sum(),
    "eksik_yuzde": (df.isna().mean() * 100).round(1),
    "benzersiz": df.nunique(),
})
tekrar = df.duplicated().sum()

print(profil.sort_values("eksik_yuzde", ascending=False))
print(f"Duplicate rows: {tekrar}")
print(df.describe(include="all").T)

3. Handle missing values

The method depends on what the column means and why the gap is there. A missing date and a missing free-text note are not the same problem.

eksik_ozeti = pd.DataFrame({
    "sayi": df.isna().sum(),
    "yuzde": (df.isna().mean() * 100).round(1),
}).query("sayi > 0").sort_values("yuzde", ascending=False)

df["bolge"] = df["bolge"].fillna("Unknown")

Keep a written record of which column you filled and why, and never overwrite the raw data. The person asking “why are there no blanks in this column?” six months from now will most likely be you.

4. Descriptive statistics

Read the mean and standard deviation together with the median and quartiles. On skewed distributions — and sales data is almost always skewed — the mean alone misleads.

sayisal = df[["adet", "birim_fiyat"]]
ozet = sayisal.describe().T
ozet["medyan"] = sayisal.median()

df["ciro"] = df["adet"] * df["birim_fiyat"]
print(ozet.round(2), df["ciro"].sum())

The revenue definition here is deliberately plain: quantity × unit price. In a real project, discounts, tax, returns and currency conversion all enter that formula — and each one changes the number. Settle the definition with the business side once, and write it into the code as a comment.

5. Grouped analysis

Produce several measures at once by region or product. Totals and averages answer different questions: one describes volume, the other the typical order size.

bolge = (df.groupby("bolge", dropna=False)
    .agg(siparis=("ciro", "size"),
         toplam_adet=("adet", "sum"),
         toplam_ciro=("ciro", "sum"),
         ort_ciro=("ciro", "mean"))
    .sort_values("toplam_ciro", ascending=False))

aylik = (df.set_index("siparis_tarihi")["ciro"]
    .resample("MS").sum().rename("ciro").reset_index())

Writing dropna=False matters: the default drops rows with a missing region from the grouping entirely, and the total revenue quietly shrinks. resample("MS") aggregates by month start and does not skip the months in between: a month with no orders enters the series as 0, not as a missing row. That is usually what you want, but on a chart “no sales that month” and “no data that month” look identical — if you need to tell them apart, check the record count per month separately.

6. Produce the charts

Let the title and axis labels explain what is being measured. Saving and closing the figure keeps memory under control in long batch jobs.

import matplotlib.pyplot as plt
import seaborn as sns
from pathlib import Path

sns.set_theme(style="whitegrid", palette="viridis")
fig, ax = plt.subplots(figsize=(8, 4))

sns.barplot(data=bolge.reset_index(), x="bolge", y="toplam_ciro", ax=ax)
ax.set(title="Revenue by region", xlabel="Region", ylabel="Revenue (TRY)")

Path("cikti").mkdir(exist_ok=True)
fig.tight_layout()
fig.savefig("cikti/bolge_cirosu.png", dpi=160)
plt.close(fig)

Put the unit in the axis label, keep labels legible, and never let colour alone carry the information. Colour blindness aside, the chart still has to be readable when the report is printed in black and white.

7. Reusable functions

Small, single-purpose functions make it easy to move the analysis between files. The real gain is not shorter code; it is preventing the same calculation from being done two different ways in two places.

def sayisal_ozet(veri: pd.DataFrame) -> pd.DataFrame:
    """Return a summary of the numeric columns plus missing counts."""
    sayilar = veri.select_dtypes(include="number")
    if sayilar.empty:
        raise ValueError("No numeric columns.")
    sonuc = sayilar.describe().T
    sonuc["eksik"] = sayilar.isna().sum()
    return sonuc.round(2)

Parameterise file paths, column names and thresholds instead of burying them inside the function. On unexpected input, raise an explanatory error rather than silently returning an empty result — like the ValueError above.

8. Export the report

Write tables to separate Excel sheets and charts to PNG files. Alongside them, record the source, the period, the filters applied and the methodology notes.

cikti = Path("cikti")
cikti.mkdir(parents=True, exist_ok=True)

with pd.ExcelWriter(cikti / "veri_analiz_raporu.xlsx") as yazar:
    profil.to_excel(yazar, sheet_name="Kalite")
    ozet.to_excel(yazar, sheet_name="Istatistik")
    bolge.to_excel(yazar, sheet_name="Bolge")
    aylik.to_excel(yazar, sheet_name="Aylik", index=False)

Put the reporting period in the file name; ten files all called rapor.xlsx help nobody. Make the script runnable with a single command and pass the input and output paths in from outside.

A workflow checklist

Start small, then harden it

Begin with a small flow that only reads the file and produces a quality profile and a short summary report. Once you are confident it works, move the repeating steps into functions and strengthen the schema and threshold checks.

The reason for that order is simple: trying to build a perfect automation from the start usually ends in a script that is never finished. A small working flow starts saving time on day one, and from then on every new file is examined by the same criteria with less manual effort.