A High Fidelity Synthetic IoT Dataset for Smart Waste Management (10 Million Records)

This dataset provides a one-year, context-aware synthetic IoT benchmark for smart waste-management research in Dhaka, Bangladesh. It represents 5,000 persistent synthetic smart-bin digital twins monitored hourly from 1 January to 31 December 2025, yielding exactly 43,800,000 hourly observations. The simulation incorporates land-use-specific activity, synthetic catchment population, population density, institutional operating status, market and recreation patterns, source-specific waste composition, informal material recovery, seasonal environmental behaviour, cultural-event scenarios, collection accessibility, overflow dynamics, and IoT sensor behaviour. The simulated urban contexts include residential and high-density settlements, fish markets, vegetable markets, general markets, commercial areas, restaurant and food zones, educational institutions, offices, garment factories, parks, playgrounds, recreation areas, and informal settlements. Waste composition is represented using source-dependent components rather than a single generic waste category. Waste mass and volume are modeled separately, enabling composition-dependent bin-filling and overflow behaviour. Contextual temporal patterns include institutional closure effects, market activity periods, recreation-area weekend behaviour, Friday-related activity, Ramadan and Eid scenarios, and annual weather seasonality. IoT-related variables include true and sensor-observed fill states, waste-generation and composition variables, collection and accessibility variables, overflow-related variables, battery telemetry, controlled sensor noise, packet-loss indicators, and anomaly indicators. The dataset is entirely synthetic and scenario-based. It is intended as a reusable urban digital-twin benchmark and should not be interpreted as field-observed municipal smart-bin measurements from Dhaka. The public release is distributed in Parquet format. The deposited archive contains the partitioned representation of the complete 43.8-million-observation dataset, and all partitions together constitute one logical dataset.