🔍 Crime Pattern Analysis Dashboard

Exploratory data analysis of ~1 million LAPD crime records (2020–2024), delivered as a reusable Python pipeline + interactive Plotly dashboard.

---

📌 Project Overview

This project analyses five years of reported crime in Los Angeles to answer three questions:

1. When does crime happen? (year / month / weekday / hour patterns)
2. Where does it concentrate? (LAPD areas, streets, geographic hot spots)
3. Against whom, and how? (victim demographics, weapons, premises)

It is deliberately structured like a real analytics deliverable rather than a single-notebook script: all logic lives in a tested src/ package (loader → cleaner → analyzer → visualizer → insights), and the notebook only orchestrates and narrates. The executive report in reports/ is generated automatically from the data — no hard-coded numbers.

🏗️ Architecture

                ┌──────────────┐
 raw CSV ─────▶ │ data_loader  │  schema-aware load, overview, dtypes
                └──────┬───────┘
                       ▼
                ┌──────────────┐
                │   cleaner    │  8-step pipeline: dedupe, dates, HHMM time,
                └──────┬───────┘  placeholder ages, imputation, features
                       ▼
        ┌──────────────┴──────────────┐
        ▼                             ▼
 ┌──────────────┐              ┌──────────────┐
 │   analyzer   │ ───────────▶ │  visualizer  │  24 themed Plotly/Matplotlib
 └──────┬───────┘  tidy frames └──────────────┘  chart builders
        ▼
 ┌──────────────┐
 │   insights   │  live metrics → key findings → executive report (md)
 └──────────────┘

Computation (analyzer) and presentation (visualizer) are strictly separated — every chart builder receives a tidy DataFrame and returns a figure.

📊 Dataset

Source	Crime Data from 2020 to 2024 — LA Open Data Portal
Size	~1M rows × 28 columns (~250 MB CSV)
Grain	One row per reported crime incident
Key fields	date/time occurred, LAPD area, crime code & description, victim age/sex/descent, weapon, premise, status, lat/lon
Status	Frozen — LAPD migrated to NIBRS reporting in 2024, so this dataset is stable historical reference data

The raw CSV is not committed to the repo (see .gitignore). Download it from the portal and place it at data/raw/crime_data_la.csv.

✨ Features

• Transparent 8-step cleaning pipeline with a logged audit trail (duplicates removed, placeholder ages nulled, HHMM times validated, Null-Island coordinates masked, structurally-missing values encoded explicitly)
• Feature engineering: year, month, weekday, hour, time-of-day buckets, weekend flag, 9-category crime taxonomy (keyword rules over ~140 raw offense codes), reporting-delay days
• 20+ analyses: temporal trends at four granularities, region comparison, victim demographics, weapon & premise rankings, correlation matrix, IQR outlier detection at record and city level
• 24 professional visualizations: bar, pie/donut, line, area, histogram, box, scatter, heatmaps, treemap, sunburst, geographic scatter map — all consistently themed
• Interactive dashboard with dropdown filters (crime category, LAPD area) that work even in static HTML exports
• Auto-generated executive report (reports/crime_analysis_report.md) recomputed live from the data
• Data-artifact awareness: noon default times, first-of-month default dates, and the 2024 NIBRS-migration cutoff are identified and handled rather than misread as trends

🛠️ Technology Stack

Layer	Tools
Language	Python 3.10+
Data wrangling	Pandas, NumPy
Visualization	Plotly (interactive), Matplotlib (static)
Environment	Jupyter Notebook / VS Code
Storage	CSV in, Parquet out (pyarrow)

📁 Project Structure

Crime-Pattern-Analysis-Dashboard/
│
├── data/
│   ├── raw/                  # place crime_data_la.csv here (git-ignored)
│   └── processed/            # cleaned parquet written by the pipeline
│
├── notebooks/
│   └── Crime_Pattern_Analysis.ipynb   # main analysis (orchestration + narrative)
│
├── src/
│   ├── utils.py              # paths, logging, LAPD code mappings, category rules
│   ├── data_loader.py        # schema-aware loading + dataset overviews
│   ├── cleaner.py            # 8-step cleaning & feature-engineering pipeline
│   ├── analyzer.py           # 20 aggregation / statistics functions
│   ├── visualizer.py         # 24 themed chart builders
│   └── insights.py           # live metrics, findings, executive report
│
├── reports/                  # auto-generated executive report
├── images/                   # exported chart PNGs
├── README.md
├── requirements.txt
├── LICENSE
└── .gitignore

🚀 Installation & How to Run

# 1. Clone
git clone https://github.com/deepthishivani/Crime-Pattern-Analysis-Dashboard.git
cd Crime-Pattern-Analysis-Dashboard

# 2. Virtual environment
python -m venv venv
venv\Scripts\activate          # Windows
# source venv/bin/activate     # macOS / Linux

# 3. Dependencies
pip install -r requirements.txt

# 4. Dataset
#    Download "Crime Data from 2020 to 2024" from
#    https://data.lacity.org (dataset id 2nrs-mtv8) → Export → CSV
#    Save it as: data/raw/crime_data_la.csv

# 5. Run
jupyter notebook notebooks/Crime_Pattern_Analysis.ipynb
#    (or open the folder in VS Code and run the notebook there)

Run all cells top-to-bottom. The pipeline logs every cleaning step, writes the processed dataset to data/processed/, exports charts to images/, and generates reports/crime_analysis_report.md.

📈 Visualizations

Category	Charts
Temporal	yearly bars with YoY %, monthly time series + rolling average, seasonality bars, weekday bars, hourly area chart, weekday×hour heatmap
Composition	category donut, top-15 crimes bar, treemap (category→offense), sunburst (category→time-of-day)
Regional	area ranking colored by dominant category, area×year heatmap, top street addresses, interactive geographic scatter map
Demographics	age histogram, age-by-category box plots, sex donut, descent bars
Statistical	annotated correlation heatmap, daily-volume outlier scatter with IQR bounds, age-vs-reporting-delay scatter
Dashboard	2×2 overview + dropdown-filtered interactive monthly trend

Screenshots

🔑 Key Insights (sample)

• Property crime (theft/burglary + vehicle crime) dominates the composition; vehicle theft is the single most reported offense in the city.
• Crime intensity peaks Friday–Saturday evenings, with weekend late-nights staying hot while weekday nights cool off — a direct patrol-scheduling insight.
• The exact-noon spike in hourly data and first-of-month spikes in daily data are LAPD default-value recording artifacts, not real patterns — identified and flagged rather than misreported.
• The steep 2024 decline is a system-migration artifact (NIBRS transition), demonstrated in the analysis rather than misread as a crime drop.
• Crime is hyper-concentrated: a handful of LAPD areas and even individual street blocks account for a disproportionate share of all incidents, and the concentration is stable across all five years.

(The full, always-current list is generated in reports/crime_analysis_report.md.)

🔮 Future Improvements

1. Census joins → per-capita crime rates by area and demographic group
2. Extend pipeline to LAPD's post-2024 NIBRS offense/victim datasets
3. Time-series forecasting (SARIMA / Prophet) per crime category
4. Streamlit/Dash deployment of the dashboard outside Jupyter
5. DBSCAN geospatial clustering to formalize hot-spot detection
6. Unit-test suite (pytest) over the cleaning pipeline

📄 License

Released under the MIT License.

🙋‍♀️ Author

Deepthi Shivani — B.Tech CSE, SRM Institute of Science and Technology GitHub: @deepthishivani