This project is a full end-to-end data analysis of the Israeli used car market, based on real listings scraped from Internet. The dataset started as raw, Hebrew-language listings and was cleaned, translated, and explored to uncover patterns in pricing, manufacturer popularity, vehicle features, and key correlations.
Data Cleaning Pipeline
1
Raw Data Loading: Loaded 21,350 rows from a CSV scraped from Internet (Feb 2026). Each row represented a single used car listing with fields including car name (in Hebrew), year, mileage, price, transmission, motor type, color, body type, and more.
2
Translation from Hebrew to English: All categorical text fields were in Hebrew. Custom translator modules were used to convert car names, motor types (e.g. בנזין → Gasoline), ownership types (e.g. פרטית → Private), colors, and body types into English for standardized analysis.
3
Dropping Invalid Rows: Rows where the car name was missing or empty were identified and removed using
dropna() on the car column. This eliminated ~2,400 rows where the scraper had returned incomplete records.
4
Removing Unknown Manufacturers: Rows belonging to rare or unrecognized Hebrew manufacturer names (113 rows across ~30 brands) that couldn't be reliably translated were dropped to keep the dataset clean and consistent.
5
Handling Missing Prices: Listings with the Hebrew placeholder "לא צוין מחיר" (price not specified) were replaced with
NaN. Missing prices were then imputed using the mean price per manufacturer + model + year group, preserving realistic price estimates.
6
Fixing Fuel Consumption: The
fuel_consumption column contained mixed types and missing values. It was converted to numeric, then missing values were filled using a cascade strategy: (1) mean by manufacturer + year + engine volume, (2) mean by manufacturer + year, (3) mean by manufacturer + motor type, and finally (4) the overall dataset mean. Electric cars were assigned 0 consumption.
7
Fixing Engine Volume: Engine volume was stored as an object type with commas and non-numeric characters. It was extracted using regex, converted to float, and missing values were imputed using the mean per manufacturer + model + year. Electric vehicles were assigned 0.
8
Handling Dates: Both
test_date and on_street_date had partial missing values. A cross-fill strategy was applied: if one date was missing, it was inferred from the other. Remaining gaps were filled using the car's model year. Dates were then converted to proper datetime objects.
9
Feature Engineering: Added a
z_price column (price z-score for outlier detection), an is_electric binary flag, and a price_category column using quantile-based binning into three tiers: cheap, medium, and expensive.
10
Final Export: The cleaned DataFrame (18,954 rows, 17+ columns, zero nulls) was saved to
cleaned_full_cars_data_v4.csv in UTF-8 encoding and used as the input for all exploratory analysis.Project Highlights
This project demonstrates a complete data science workflow: from raw web-scraped data with mixed Hebrew/English text, through multi-step cleaning and imputation, to rich exploratory visualizations. It covers distribution analysis, manufacturer comparisons, price modelling, feature correlation, and pairwise relationship analysis — all implemented in Python using Pandas, Matplotlib, and Seaborn.
Analysis & Visualizations
