Almaty Real Estate & Housing Price Analysis

Project Objectives

1. Exploratory Data Analysis (EDA): Evaluate price per square meter across administrative districts, building materials, and construction periods.
2. Data Cleaning & Outlier Removal: Filter out duplicate listings, corrupt floor numbers, and extreme pricing outliers.
3. Predictive Valuation Modeling: Train machine learning algorithms (Random Forest, LightGBM) to forecast market apartment prices.

Project Workflow (Flowchart)

The diagram below maps how this site was built — from the raw Kaggle dataset all the way to the interactive charts and the predictive model shown on this page.

Raw Data almaty_apartments_with_id.csv — ~12,000 listings

Data Cleaning Remove duplicates, fix floor numbers, drop price outliers

Exploratory Data Analysis Price per m² by district, building type, and room count

Hypothesis Testing ANOVA · Two-sample T-test · Shapiro-Wilk

Predictive Modeling Random Forest & LightGBM regression (R² ≈ 0.82)

Visualization & Insights Interactive charts and key findings published on this site

Statistical Hypothesis Testing

1. One-Way ANOVA: District Price Disparity

Null Hypothesis (H₀): Mean apartment price per square meter is equal across all administrative districts of Almaty.

Alternative Hypothesis (H₁): At least one district has a significantly different mean price per square meter.

Test Result: p-value < 0.001 (Reject H₀). There is a statistically significant price premium in Medeu and Bostandyk districts.

2. Two-Sample T-Test: Monolithic vs. Panel Buildings

Null Hypothesis (H₀): There is no difference in the average price per square meter between monolithic and panel buildings.

Alternative Hypothesis (H₁): Monolithic buildings cost significantly more per square meter than panel buildings.

Test Result: p-value < 0.01 (Reject H₀). Monolithic structures carry an average validated premium of 22% over panel units.

3. Shapiro-Wilk Test: Price Normality

Null Hypothesis (H₀): Real estate prices follow a normal distribution.

Alternative Hypothesis (H₁): Real estate prices deviate from a normal distribution.

Test Result: p-value < 0.001 (Reject H₀). The price distribution is strongly right-skewed; logarithmic transformation is required for linear modeling.

Dataset Schema (almaty_apartments_with_id.csv)

Feature	Data Type	Description
`id`	Numerical (Integer)	Unique identifier of the listing
`district`	Categorical (String)	Administrative district (e.g., Medeu, Almaly, Bostandyk)
`rooms`	Numerical (Integer)	Total number of rooms
`area`	Numerical (Float, m²)	Total area in square meters
`floor` / `max_floor`	Numerical (Integer)	Apartment floor level and total building storeys
`building_type`	Categorical (String)	Structural material (Monolith, Brick, Panel)
`year_built`	Numerical (Integer)	Year of building completion
`price`	Numerical (Float, KZT)	Target variable: Advertised asking price

Dataset Visualizations

Figure 1: Median Price per m² by District (KZT)

Shows significant geographic disparity between upper foothill districts and northern peripheral districts.

Figure 2: Distribution of Listings by Building Type

Proportion of available properties across modern monolithic, Soviet panel, and brick buildings.

Figure 3: Median Total Asking Price by Room Count

Total apartment price progression based on the number of rooms.

Key Findings

Location is Primary: Medeu and Bostandyk properties trade at a 35% to 50% premium relative to Turksib and Alatau districts.
Structural Longevity: Modern monolithic construction built after 2010 demonstrates significantly higher value retention than Soviet housing stock.
Predictive Baseline: An ensemble Gradient Boosting Regressor achieved an R² score of ~0.82 on the test split.