On June 10, 1968, the United States Supreme Court issued a decision on Terry v. Ohio case, in which an officer in plain clothes had conducted a stop-and-frisk search on three men he believed were about to rob a store. The officer found weapons on two of the men and convicted them of carrying concealed weapons. The two men had appealed the case, saying that the evidence used to convict them in the trial was found during an unreasonable search. The U.S. Supreme Court decided that police stop-and-frisks were not in violation of the Fourth Amendment of the U.S. Constitution, which prohibits unreasonable searches and seizures. This case introduced the notion of “reasonable suspicion,” which allows officers to detain a suspect without any clear evidence required for an arrest.
Terry Stops were stops of suspicious drivers conducted by Seattle police officers. According to the Seattle government website, “This data represents records of police-reported stops under Terry v. Ohio, 392 U.S. 1 (1968). — Each record contains perceived demographics of the subject, as reported by the officer making the stop and officer demographics as reported to the Seattle Police Department, for employment purposes”.
The data was obtained from the www.data.gov City of Seattle website. This data represents records of police-reported stops under Terry v. Ohio, 392 U.S. 1 (1968). Each row represents a unique stop. Each record contains the perceived demographics of the subject, as reported by the officer making the stop. The Seattle Police Department documented the officer demographics for employment purposes. Where available, data elements from the associated Computer Aided Dispatch (CAD) event (e.g., Call Type, Initial Call Type, Final Call Type) are included. For this data analysis, I am looking to predict arrests using features in the dataset.
Exploratory Data Analysis
The following are graphics obtained after data scrubbing the dataset and conducting some exploratory data analysis using Numpy, Pandas, Matplotlib, and Seaborn Python packages/libraries. I proposed some questions to analyze the dataset better.
Which Race is Most Stopped?
What demographic of officers perform the most stops?
What is the Distribution of the Officer’s Ages?
What Age Groups are Most Stopped? Are they Carrying Weapons?
What Subject Races are Most Stopped and are they Carry Weapons?
After cleaning the provided Terry Stops dataset, I split the data 75/25 using the sklearn’s train_test_split package. The data containing continuous values were normalized using sklearn’s StandardScaler() package. The data containing categorical values were one-hot encoded. Using the imblearn.over_sampling SMOTE() function, the imbalanced data was balanced by increasing the minority class. I built a custom classifier with sklearn’s BaseEstimator with ClfSwitcher to pass in any classifier and parameters for each classifier. This custom classifier was used along with a pipeline and GridSearchCV. The classifiers used included: KNeighborsClassifier(), RandomForestClassifier(), AdaBoostClassifier(), and GradientBoostingClassifier().
GradientBoostingClassifier() had the best model performance. This classifier showed to have a training accuracy score of 0.945 and a testing accuracy score of 0.820. I found the most important features in the dataset to be ‘frisk,’ ‘officer_yob,’ ‘stop_resolution_Arrest.’
A majority of Terry Stop subjects were White, immediately followed by Black/African-Americans. According to Seattle’s government website, Seattle’s African-American population is 6.8% of the whole population compared to Seattle’s White population that make up 68% of the whole. However, African-Americans make about 1/3 of Terry Stops. Across all races, the most stopped subjects were those belonging to the age group 26–35. This age group made up a third of the dataset and are also more likely to carry weapons on their person. As for officer demographics, a majority of officers were white men.
For the government of Seattle and the Seattle Police Department to better serve their citizens, I would recommend the following. First Seattle Police department should be hiring more officers belonging to minority communities and women as well. A majority of the Seattle PD are white men, which could cause problems in policing minority communities. Women officers can bring a new perspective to the police force, which would prove to be a good change. I would recommend Seattle PD to reallocate their weapons budget to advancing health care plans to include mental health services like therapy so that officers can better understand themselves and any biases they may have towards specific communities. Most Terry Stop subjects do not have weapons, so there is no need for officers to have so much military aid. With more focus on bettering police practices and the police themselves, there will be fewer lawsuits against officers for abuse of power, which saves Seattle a lot of money.
This dataset contained a lot of NaN or incomplete information, so more data collection should be made before this analysis can be 100% approved. Many data columns had a majority of the data missing, so I did not use them in my study even though they could have provided some interesting insights. I would want more data collection to occur in minority communities because I am sure the Terry Stop data would be a lot different from this dataset.