telco_churn_anonymized_dataset_italy

This dataset supports the research presented in the paper:"[FILL IN: Article Title]", submitted to [FILL IN: Journal]. Overview--------An anonymized, semester-aggregated dataset of customer records from a real-worldItalian telecommunications company, curated for churn prediction research.The dataset covers multiple semesters and contains over 16 million records acrossapproximately 4.5 million unique customers. The primary prediction target is will_churn: a binary variable indicating whethera customer discontinued their service in the semester following the observation period. Dataset Structure-----------------The data is organized by semester (e.g., '2023_0', '2023_1', '2024_0', '2024_1').Due to upload size constraints, the dataset is split into four files, one per semester.To use with the provided training scripts, merge them into a single CSV namedtelco-dataset-anonim.csv, or modify the scripts to load files individually. Key Features------------- wifi_offer_name: Wi-Fi subscription plan name (categorical, anonymized)- wifi_fee_sem_k: Wi-Fi revenue for the semester (thousands of EUR)- mobile_fee_sem_k: Mobile revenue for the semester (thousands of EUR)- n_mobile_offers_sem: Number of active mobile subscriptions in the semester- payment_method: Customer payment method (categorical, anonymized)- tenure_norm: Normalized customer tenure- user_engagement_score: Composite engagement metric- total_pendings_sem_k: Total pending payments for the semester (thousands of EUR)- age_norm: Normalized customer age- cap_extracted: Postal code area (anonymized)- will_churn: Binary target variable (1 = churned in next semester, 0 = retained)- customer_code: Anonymized customer identifier (SHA-256 hash)- semester: Observation period label Anonymization-------------All direct and indirect personal identifiers have been anonymized in compliancewith GDPR. Customer codes were transformed using SHA-256 cryptographic hashing.Categorical variables containing potentially identifying information weregeneralized or encoded. No personally identifiable information (PII) or rawbilling details are present in this dataset. Pre-processing--------------Missing values in categorical columns were imputed with the string "Missing".The 'vat' field (VAT registration status) was treated as a boolean string.Numerical features were normalized. The dataset is ready for direct use withthe companion code repository. Companion Code--------------https://github.com/franc96/telco-churn-prediction-ml License-------Creative Commons Attribution 4.0 International (CC BY 4.0)