An Optimized Machine Learning and Big Data Approach to Crime Detection

Table 1

Summary of related works.

References	Dataset	Methods used	Evaluation metrics	Limitations
Navalgund and K. (2018) [4]	YouTube and Google	VGGNet -19	Accuracy, recall, F1-score and support	Detection of crime hotspots and probability of occurrences not included.
Younghyun Lee et al. (2011) [5]	Real-time elevator data collected using surveillance camera of pixels	Violent frame detector, motion vector extraction, and foreground segmentation	Detection rate, no. of people in the elevator, false-positive rates (FPR)	Includes only detection but not prediction or probabilities of occurrence results
Nakib et al. (2018) [6]	Real-time data	Softmax regression model, CNN	Accuracy	The size of the dataset was relatively small.; The model was not evaluated against the other classical models.
Ranjan et al. (2018) [7]	Image collected from various internet sources and then morphed to test the methods	SVM and ANN	Accuracy, sensitivity and specificity	Comparison of the results with other traditional approaches were not included.; Availability of larger dataset also is a challenge
Vynokurova et al. (2020) [8]	Real-time dataset	SVM and random forest-based hybrid approach	Accuracy	Comparison of the results with other traditional approaches were not included.; Availability of larger dataset also is a challenge