Crime Hotspots Forecasting via Deep Learning Methodology
Location
Hager-Lubbers Exhibition Hall
Description
PURPOSE: Crime poses a significant social issue in the United States, threatening public safety and impacting the economy profoundly. Effectively predicting and understanding patterns of criminal activity is crucial for identifying future high-risk areas, commonly known as "crime hot spots." Accurate identification of these areas allows law enforcement agencies to allocate resources more effectively and respond efficiently to incidents. The advent of sophisticated data collection and storage technologies has resulted in the accumulation of a large volume of spatial and temporal data related to criminal activities. The challenge lies in harnessing this vast spatial-temporal information to precisely forecast regional crime rates, essential for proactive crime prevention. PROCEDURE: To precisely predict these hot spots, we leveraged machine learning technology and employed several deep learning neural network architectures to learn the patterns of historical call-for-service data from the Portland, Oregon Police Bureau. After preprocessing the raw data, it was then forwarded to models with training and testing sets. We left one set of data out from the model to ensure predictions were made without learning from this set. OUTCOME: After comparing the predictions from each model, a new model that incorporates both Convolutional Neural Networks and Long Short-Term Memory model outperformed among the models we trained. IMPACT: Utilizing the hybrid model may be beneficial in developing a more accurate crime hotspot forecasting model. We learned the importance of data preprocessing, which significantly enhances the quality of predictions by ensuring the data is appropriately formatted and cleaned, we gained experience in understanding the type of data used and how different parameters can impact the results of prediction.
Crime Hotspots Forecasting via Deep Learning Methodology
Hager-Lubbers Exhibition Hall
PURPOSE: Crime poses a significant social issue in the United States, threatening public safety and impacting the economy profoundly. Effectively predicting and understanding patterns of criminal activity is crucial for identifying future high-risk areas, commonly known as "crime hot spots." Accurate identification of these areas allows law enforcement agencies to allocate resources more effectively and respond efficiently to incidents. The advent of sophisticated data collection and storage technologies has resulted in the accumulation of a large volume of spatial and temporal data related to criminal activities. The challenge lies in harnessing this vast spatial-temporal information to precisely forecast regional crime rates, essential for proactive crime prevention. PROCEDURE: To precisely predict these hot spots, we leveraged machine learning technology and employed several deep learning neural network architectures to learn the patterns of historical call-for-service data from the Portland, Oregon Police Bureau. After preprocessing the raw data, it was then forwarded to models with training and testing sets. We left one set of data out from the model to ensure predictions were made without learning from this set. OUTCOME: After comparing the predictions from each model, a new model that incorporates both Convolutional Neural Networks and Long Short-Term Memory model outperformed among the models we trained. IMPACT: Utilizing the hybrid model may be beneficial in developing a more accurate crime hotspot forecasting model. We learned the importance of data preprocessing, which significantly enhances the quality of predictions by ensuring the data is appropriately formatted and cleaned, we gained experience in understanding the type of data used and how different parameters can impact the results of prediction.