International Journal of Engineering and Management Research

1. Introduction

According to the 2024 World Drug Report released by the United Nations Office on Drugs and Crime (UNODC), more and more people use drugs for medical and non-medical purpose, reaching 292 million people worldwide in 2022, a 20 per cent increase over 10 years. Another online source by the National Center for Drug Abuse Statistics (NCDAS) reported that, in USA alone, over a half of people 12 and older have used illicit drugs at least once, 700,000 people died from drug overdose since 2000, and the federal government spent $35 billion in 2020 alone for drug control and prevention.

To exacerbate the current drug-related problems that many countries face, the emergence of new synthetic drugs with mass manufacturing facility, a record high level of drug trafficking, and liberal legalization of drugs are negatively impacting people’s health and well-being. For example, cannabis has been the most commonly used drug (228 million users) and its dominance is likely to continue due to the recent legalization of its production and sale for non-medical use in Canada and over a half of jurisdictions in the US. As expected in advance, more people (especially among young adults) in these two countries suffered from over (or even regular) use of cannabis, leading to psychiatric disorders and attempted suicide since its legalization (NCDAS, 2020).

It is also noteworthy that more than 16 million people in US misuse prescriptions for painkillers, opioids and sedatives in a year, which leads to a high rate of drug abuse, addiction and overdose (NCDAS, 2020). At the same time, a growing number of people are exploring the medical use of psychotropic substances to treat some mental health disorders without clinical and scientific guidelines. Note that psychotropic substances are chemical substances that can affect people’s cognitive ability (consciousness and cognition), emotion (perception and mood), or behavior.

The afore-mentioned prevalence of drug overdose and the misconception on psychotropic substances naturally lead to drug use disorders, drug offences and environmental harms. For example, about 64 million people out of 292 million who used drugs worldwide are estimated to suffer from drug use disorders, only less than 10 per cent of them is currently in treatment (UNODC, 2024).

That is, it is very possible that geographical locations associated with frequent illicit drugs use or possession of equipment to consume illicit drugs can be different from locations of frequent traffic violations under influence of drugs or alcohol beverages. Once such locations are tagged for each kind of drug crimes, local law enforcement office may reconfigure daily patrol paths with officers with more experiences and appropriate equipment for specific kinds of drug crimes. Such information will be also very valuable for public use to avoid dangerous areas for safety reasons.

In this paper, we intend to collect, analyze, and visualize crime data sets collected through incident-based reporting (IBR) by local police and sheriff’s offices from 2011 through 2019 in a local community. In our definition of drug-related crimes, three types of drug-related crimes are included. The first type is Drug category that contains all crime incidents related to (suspicious) drug use and possession of equipment to consume drugs. The second type included is Liquor category such as alcohol offense and intoxication. The reason that we include alcohol offense and intoxication as drug-related crimes is that very high levels of alcohol in the body will significantly impair areas of human brain related to cognition, emotion, and behavior, which is very similar to symptoms of drug overdose. The last type considered is Traffic violation associated with driving under influence (Traffic-DUI) of drugs and alcohol.

Once we finalize the data sets of drug-related crimes, we like to review and visualize temporal and seasonal patterns of crime incidents. For advanced analysis, we intend to find a natural grouping of the geographic locations of crime incidents using clustering analysis. To this end, we employ a density-based clustering algorithm (DBSCAN), which will find geospatial clusters based on longitude and latitude information of crime locations. Finally, we will profile each cluster based on prevalent crime types within the clustered location. The visualization of such clusters on a map will make it easy for residents and public safety officers to recognize hot spots of drug-related crimes.

2. Literature Review

In this section, we like to briefly review what clustering analysis is and introduce several well-known clustering algorithms.

The EM algorithm is known to outperform other distance-based clustering algorithm (e.g., K-means) because of its applicability on both continuous and categorical data sets (Meila & Heckerman, 1998). In addition, the EM algorithm is very unique in a sense that it assigns the likelihood of the data point being in those clusters and hence each data point may belong to multiple clusters (i.e., soft clustering) while most other clustering algorithms (e.g., K-Means) assign each data point to only one cluster (i.e., hard clustering).

Another clustering algorithm group is hierarchical clustering algorithm that returns a hierarchical series of nested clusters called Dendrogram. In hierarchical clustering algorithm, there are two algorithms: Agglomerative clustering and Divisive clustering. In Agglomerative clustering, each data point forms its own cluster and, at every iteration, two nearest clusters are merged until one cluster is formed (Griffiths et al., 1984). In contrast, Divisive clustering initially assumes that all data points belong to a single cluster and recursively divide dissimilar data points into smaller clusters until each data point becomes its own cluster. While hierarchical clustering shows the hierarchical structure of the data sets, the computational cost in terms of CPU and memory requirement is relatively high and the resulting outcomes is often dependent on initial conditions for implementations (Murtagh & Contrera, 2012).

The last category of clustering algorithms is density-based clustering algorithm. In this category, the density-based spatial clustering of applications with noise (DBSCAN) is one of the most well-known and cited algorithms. The DBSCAN uses density measure as clustering criterion, and hence identifies data points with many nearby points (= a high dense region) as clusters and points in low density regions as outliers.

The key idea of the DBSCAN is that for each point to belong to a cluster, the neighborhood of a given radius (controlled by eps parameter) has to contain at least a minimum number of points (controlled by minPts). Therefore, the DBSCAN identifies each data point as a core point if it has more minPts within eps, a border point if it has fewer than minPts within eps but it is in the neighborhood of a core point, or a noise point if it is not a core point or a border point (Ester et al., 1996).

Table 1: List of Variables

During our preliminary inspection on the data set, we found few erratic records (e.g., nine crime records whose incident year of crime was earlier than 2011), which we removed from our analysis. While we found it weird to observe only two records in the year 2013, we decided to keep them because they had complete information and it is not our main goal to study yearly trends of crimes.

The brief introduction to the local community, Cache County in Utah, USA, is necessary to understand and complete geospatial specific analysis. The authors acknowledge that the following information about Cache County is borrowed from two Web sites: Cache County official site (https://www.cachecounty.gov/) and Wikipedia (https://en.wikipedia.org/wiki/Cache_County,_Utah) solely for the purpose of self-containment of this paper.

The Cache County is located in the northern region of Utah state bordering Idaho state. Historically, it was first formed in 1856 and currently has 13 cities such as Logan (county seat and largest city), Hyrum, North Logan, Smithfield, Nibley, Wellsville, Amalga, and so on. It also hosts two public universities, Utah State University (USU) and Bridgerland Technical College, in Logan.

The population in Cache County in 2023 was estimated to be 142,393 and male and female groups almost evenly represent the population. However, the largest population came from the age group of [18, 64] (60.6%) followed by young group (<18, 29.3%) and senior group (>=65, 10.2%).

Our analysis revealed that in terms of parent crime types, top 10 most frequently occurred parent crime types were Community Policing (43.4%) followed by Traffic (23.2%), Theft (5.6), Disorder (4.9%), Alarm (4.7%), Property Crime (4.0%), Drugs (2.5%), Pedestrian Stop (1.9%), Liquor (1.9%), and Assault (1.6%). We noted that the most frequent crime type, Community Policing, was mainly related to animal problem, citizen assistance, and welfare check, while the second most frequent crime type was related to traffic accident, traffic offense, and vehicle identification number (VIN) inspection.

4.2 Seasonal Patterns of Psychotropic Substances Crimes

In this subsection, we like to focus our analysis only on psychotropic substances crime (PSC) mainly because this research was in part initiated with the purpose of collaborating with local law enforcement offices in the city of Logan that want to gain insights of geospatial relationships and devise an actionable set of recommendations to control the increasing trend of PSC. In addition, Logan city is the county seat of Cache County and hence public safety issues in Logan garner the attentions of city council members and Cache County administrators. Therefore, we analyze PSC incidents only in Logan from now on.

Figure 2: PSC Statistics in Logan City

According to Figure 3, PSC maintained a very high level of incidents in the period of September (1,360 incidents), 2011 through April (1,221), 2012. In the academic year of 2014, incidents started to increase August, peaked in November (183) and December (550), but downed to almost zero in January, 2015. During the academic year of 2015, however, incidents steadily increased from August, sharply increased from January (140) in 2016, and peaked in April (411) in 2016. Since then, it sharply decreased but still showed relatively high incidents level in May-July (over 150 incidents). In the academic year of 2017, it mainly remained stable and low until November (21), but sharply increased and maintained a very high-level during January (352) and February (391) until May (130) in 2018. The PSC incidents during the academic year of 2018 did not show a particular seasonal pattern like in previous years. Instead, it maintained a steady level of incidents between 3 and 55 over the entire year.

We also separated the seasonal patterns of Drug, Liquor and Traffic incident type and presented them at the bottom panel of Figure 3. Overall, each of Drug, Liquor and Traffic incident type presented a similar seasonal pattern we observed from the aggregated PSC category at the top panel: peaked incidents in January through March.

From these observations, we concluded that seasonal patterns of PSC incidents were closely related to academic calendars. We partially attribute this finding to the fact that the city of Logan hosts two public universities and hence it shows typical patterns observed in many other college campus towns. That is, when a new academic year starts in late August, first-time and returning students are likely to have various social meetings and hence they are most likely to be exposed to opportunity of having drinks or even psychotropic substances, and driving back to home or dormitories under the influences of alcohol.

However, this single factor alone does not fully explain why PSC incidents are the highest from January to March. After careful thought, we attribute cold winters weather in Logan city as the second factor to the high PSC incidents in winter seasons. Geographically, the Logan city is located in the highly elevated northern region (4,534 ft or 1,382 m) and has very warm and dry summers that attract many temporary residents from southern states such as Arizona and Florida.

Figure 4 provides one additional information of which locations of the city are hot spots of PSC incidents in darker red colors. That is, locations with darker red colors indicate multiple incidents of PSC. We immediately noted that many hot spots of PSC incidents were aligned with US highway 91 (shown along the vertical center of the map in Figure 4. Many local residents refer it as the main road) that runs through from Brigham City, Utah to Idaho Falls, Idaho. This perfectly makes sense because it is a well-known fact that the success of restaurants, bars, theaters, inns and hotels, coffee shops and many other types of business is heavily dependent on influx of more residents or travelers. Therefore, they are mostly located along the highway that passes through the center of the city because it grants residents and travelers easy access with vehicles.

However, the hottest spot of PSC incidents with most locations in darker red colors was observed at the intersection of US highway 91 and Utah State Route 30 (SR-30) (shown along the horizontal center of the map in yellow in Figure 4. Many local residents refer it as the center street). With easy access through both SR-30 and US highway 91, the intersected area of two highways indeed has served as the central location of business entities and administrative offices in the city of Logan. Therefore, it is not surprising to pinpoint this area as the hottest spot of PSC incidents.

One major limitation of findings from Figure 4 is that such findings provide only macro-level understanding of PSC incidents in the city. Therefore, they are not particularly useful for administrators in the local law enforcement office to develop efficient and effective PSC prevention plans including optimal officer allocation and patrol path configuration suited for each subcategory of PSC. For example, possibly best patrol locations for police officers to stop and arrest drivers who are suspected to commit Traffic-DUI crime may not be consistent with promising locations to spot and arrest the violators of C/S Drugs or C/S Possession of Paraphernalia (Denoted as C/S Poss Para in chart labels). Note that paraphernalia refers to any equipment to produce, conceal and consume illicit drugs.

4.4 Validation and Discussion of DBSCAN Clustering

For administrators in police and sheriff office to revise their current PSC prevention plans for each

Cluster 0 in Figure 5 displays somewhat balanced distributions of four subcategories, DUI (38.0%) followed by Alcohol Offense (24.1%), Susp(ected) Drugs (20.7%) and C/S Drugs (17.2%). Cluster 0 can be also characterized with three main categories, Traffic-DUI (38.0%), Drug (37.9%) and Liquor (24.1%). Cluster 1 display a very similar distribution found in Cluster 0. Single dominant

PSC category in Cluster 1 is Traffic-DUI (46.9%) followed by Liquor-Alcohol Offense (25%) and Drugs (28.1%). In short, both Clusters 0 and 1 are commonly characterized as Traffic-DUI lead clusters while the second and third dominant categories in their clusters are different.

To inspect geospatial hot spots of Cluster 0 and Cluster 1, we plotted all data points in each cluster on a map and showed them in Figure 6. Two main observations were noted from Figure 6. First, since DBSCAN algorithm separated these two clusters, their geospatial locations were not overlapped. In particular, two hot spots in the north-west area (i.e., one near Bridgerland College and another residential area around 1800 North and 500 West) were marked for Cluster 0 while two hot spots in the north (i.e., commercial area on 1400 North) and east area (i.e., inside of USU campus near Dee Glen Smith Spectrum) were identified for Cluster 1. However, some of their hot spots were located close around the central commercial locations of the city mainly because Traffic-DUI incidents category is a dominant lead in both clusters.

Figure 6: PSC Hot Spots of Cluster 0 & 1 in Logan

Cluster 2 and Cluster 3 in Figure 4 shared a common characteristic in a sense that Drug category is a dominant category (69% and 55%, respectively) followed by Alcohol and Traffic-DUI categories with slightly different values. We also presented geospatial hot spots of Cluster 2 and Cluster 3 in Figure 7.

In addition, this location is within a walking distance from Days Inn and Japanese restaurant, thus constantly attracting many people. Three hot spots of Cluster 4-5-6 along 1400 North also attract many people due to easy access for various business entities including Walmart (one of the largest shopping mall chains in USA), numerous restaurants and Intermountain Health Hospital.

Figure 8: PSC Hot Spots of Cluster 4-6 & 7-9

One hot spot (located at 200 East and 900 North) of Cluster 7-8-9 in Figure 8 was particularly intriguing to the authors mainly because this is where a middle school is located. While it is unknown whether violators on this hot spot were students in this school, school administrators and police and sheriff department are strongly encouraged to educate students in the school and neighborhood district about the danger of using psychotropic substances because its impact on young children tends to be more detrimental and lasts longer.

5. Conclusion

This study intends to improve the public safety in a local community by locating multiple hot spots of various crimes related to the usage of psychotropic substances. In particular, this study employs DBSCAN clustering algorithm to cluster geographical locations of such crimes and profile clusters based on dominant crime types.

To this end, we validated all 10 clusters in terms of non-overlapped geospatial locations and provided insights on what geospatial characteristics and business environmental features make each identified cluster become the hot spots of psychotropic substances crimes. Note that afore-mentioned findings and discussion are based on data sets extracted only from a specific local community.

[11] Murtagh, F., & Contreras, P. (2012). Algorithms for hierarchical clustering: An overview. WIREs Data Mining and Knowledge Discovery, 2(1), 86-97.

[12] Ester, M.,Kriegel, H.-P.,Sander, J., & Xu, X. (1996). Simoudis, E., Han, J., & Fayyad, U.M. (eds.). A density-based algorithm for discovering clusters in large spatial databases with noise. Proceedings of the Second International Conference on Knowledge Discovery and Data Mining (KDD-96), pp.226–231.