Associating Depressive Symptoms in College Students with Internet Usage Using Real Internet Data

Raghavendra Kotikalapudi*, Sriram Chellappan*, Frances Montgomery+, Donald Wunsch and Karl Lutzen * Department of Computer Science + Department of Psychological Sciences Department of Computer Engineering

Information Technology Department

Missouri University of Science and Technology Rolla, Missouri 65409, USA. {rkyvb, chellaps, berner, dwunsch, kfl}@mst.edu
Abstract Depression is a serious mental health problem affecting a large population of college students. Since college students extensively use the Internet, the Psychological Sciences community is investigating associations between Depression and Internet usage. While existing studies provide insightful conclusions, the Internet usage in these studies is characterized by surveys alone, hence yielding limited insights. In this paper, we report our findings on a month long experiment conducted at Missouri University of Science and Technology on associating depressive symptoms among college students with their Internet usage using real Internet data collected in an unobtrusive and privacy preserving manner over the campus network. In our study, 216 undergraduates were surveyed for depressive symptoms using the CES-D scale. We then collected their oncampus Internet usage characterized via Cisco NetFlow data. Subsequent analysis revealed that several Internet usage features like average packets per flow, peer-to-peer (octets, packets and duration), chat octets, mail (packets and duration), ftp duration, and remote file octets exhibit a statistically significant correlation with depressive symptoms. Additionally, Mann-Whitney U-tests revealed that average packets per flow, remote file octets, chat (octets, packets and duration) and flow duration entropy have a statistically significant difference in the mean values across groups with and without depressive symptoms. To the best of our knowledge, this is the first study that associates depressive symptoms among college students with their real Internet usage. Keywords - depression, internet, mental health, college students I. Internet usage using real campus Internet data collected unobtrusively and in a privacy preserving manner. A. Internet usage as a marker for Depression Recent studies show that more than 90% of college students in the US actively use the Internet [23, 26]. While the benefits of Internet for academic learning, research, business and social networking are well known, studies conducted by the Psychological Sciences community have focused on exploring relationships between Internet use and students mental health. Studies in [3, 4, 6, 7] demonstrated that students with depressive symptoms used the Internet much more than those without symptoms. It was also shown that when the Internet was utilized for activities like shopping, depressive symptoms among students increased [5]. Excessive online video viewing [18, 19, 20], social networking [31], online gambling [9, 10], frequent visits to health-related websites [11], excessive late-night Internet use [12, 13] and online chatting [21, 22] have also been associated with depression among young people. With excessive Internet use, students replace real-life interactions with online socializing, leading to increased social isolation and anxiety in their physical environments [8]. While all of the above studies provide critical insights into understanding how Internet associates with depressive symptoms in college students, the information they convey is limited. This is because the student Internet usage was assessed by means of surveys only. In other words, students themselves reported their volume and type of Internet activity. This method presents limitations. First, the volume of collected Internet usage data is limited during surveying because peoples memories fade with time. There may be errors and social desirability bias when students report their own Internet usage. An accurate characterization of Internet usage requires representations of significantly higher dimensionality, and clearly the number of dimensions that can be captured via surveys is limited. B. Contributions of this Paper We conducted a study in 2011 for associating depressive symptoms among college students with their real Internet usage data collected unobtrusively and in a privacy preserving manner 1 at
1

INTRODUCTION

Depression is a serious mental health problem affecting a large segment of society. Particularly concerning are increasing incidences of depression among college students. In the US, surveys show that 10 - 40% of college students suffer from depressive symptoms at some point [1, 2]. Although there are effective treatments for depression, victims do not seek treatment because they do not recognize the symptoms, or are reluctant to do so [24, 25]. If left untreated, depression can cause severe health impacts such as appetite loss, sleep disorders, fatigue, anxiety and panic attacks. Depressed students also suffer from poor academics, decreased work place performance, and high dropout rates. As such, there is a critical need today for early and proactive detection of depressive symptoms. In this paper, we report our findings on a month long experiment conducted at Missouri University of Science and Technology on associating depressive symptoms among college students with their

This research was proposed to the Institutional Review Board (IRB) at Missouri S&T, and received approval under Exempt Category 4: Research involving the collection or study of existing data, documents, records, pathological specimens, or diagnostic specimens, if these sources are publicly available or if the information is recorded by the investigator in such a manner that participants cannot be identified, directly or through identifiers linked to the participants. 1

Missouri S&T. To the best of our knowledge, this is the first study to do so. The study consisted of the following steps: Participant Selection and Surveying: We recruited 216 student participants from three undergraduate classes at Missouri S&T in February 2011. The depressive symptoms of participants were quantified using the Center for Epidemiologic Studies Depression (CES-D) scale [14]. In our survey, 30% of students met the minimum CES-D criteria for exhibiting depressive symptoms, which compares well with national estimates of 10 - 40% [1, 2]. Internet Usage Feature Extraction: The Internet usage activity of participants was obtained in the form of Cisco NetFlow data collected over the Missouri S&T campus network. For each participant, we derived a number of Internet usage features divided into three broad categories. The Aggregate category captures raw aggregates of Internet usage like flows, packets, octets, durations etc. The Application usage category captures application specific Internet usage features like chatting, peer-to-peer, email, ftp, http etc. The Entropy based features captures randomness in Internet usage from the perspective of flows, octets, packets, durations etc. Statistical Analysis: Statistical Analysis was performed, which revealed that the following Internet usage features correlate with depressive symptoms: average packets per flow, peer-to-peer (octets, packets and duration), chat octets, mail (packets and duration), ftp duration, and remote file octets. Additionally, Mann-Whitney U-tests revealed that average packets per flow, remote file octets, chat (octets, packets and duration) and flow duration entropy have a statistically significant difference in the mean values across groups with and without depressive symptoms. Interpretation of Results: We derived preliminary interpretations to our findings by integrating them with existing research in Psychological Sciences on associations between depressive symptoms and Internet usage among college students.

(where participants form an interpretation of the experiments purpose and unconsciously change their behavior accordingly), the survey was titled Recent Affective Experiences Questionnaire, and additional items were embedded into the original CES-D questionnaire. Table I summarizes our participant pool.
TABLE I.

SUMMARY OF OUR PARTICIPANT POOL

Psychology CES-D 16 CES-D < 16

Computer Science Male Female Totals

120 8 128

68 20 88

54 10 64

134 18 152

To ensure privacy of participants, appropriate anonymization techniques were enforced during selection and surveying. The IT department at Missouri S&T provided unique pseudonyms for each participant, and the associations were not disclosed to the research team. Students who completed the CES-D survey did so using only their pseudonyms, which were tied to their recorded CES-D scores. The IT department remained unaware of the CES-D scores. Additionally, the IT department provided the on-campus Internet usage data indexed only by pseudonyms, hence preventing associations between real student IDs and Internet usage. The only associations available to the researchers were between Internet usage and CES-D scores. In our study, IP addresses were not processed, since the focus was on broad Internet statistics alone 2 . Also, the contents of emails, chat and ftp uploads/downloads were not recorded due to privacy considerations. III.
INTERNET DATA COLLECTION AND PREPROCESSING

C.

Paper Organization The remainder of this paper is organized as follows. Section II describes the participant selection, CES-D survey, and privacy preserving mechanisms. The Internet data collection process and all pre-processing techniques are detailed in Section III. Section IV describes Internet features extraction process. Results from statistical analysis are presented in Section V. Interpretations and applications of our findings are presented in Section VI and the paper is concluded in Section VII. II.

PARTICIPANT SELECTION AND CES-D SURVEY

In our study, the participant pool consisted of 216 undergraduate students at Missouri S&T from three classes: Psych 50 (General Psychology), CS 284 (Operating Systems) and CS 153 (Data Structures). Psych 50 is taken by students from all departments, while CS 284 and 153 are taken by students from many engineering departments. The survey was preceded by a consent form, and there was a minimum age of at least 18 years to participate. The survey was conducted in February 2011. The levels of depressive symptoms among participants were quantified with a one-time survey based on the Center for Epidemiologic Studies Depression (CES-D) scale. The CES-D scale was developed by Lenore Radloff of Utah State University and is used to measure depression levels in the general population [14]. It consists of 20 questions rated on a 4-point Likert scale. Possible scores range from 0 to 60, with higher scores indicating greater levels of depressive symptoms. In general, a score of 16 or above on the CES-D scale is considered indicative of depressive symptoms. The CES-D scale is widely used and has been extensively tested and validated. It has been shown to be reliable when testing adolescents in high schools and colleges [15, 16]. In order to minimize demand characteristics

The main source of Internet Usage data for this study was NetFlow. Cisco NetFlow technology is a protocol for collecting IP traffic information and is popular. NetFlow data consists of several flows. In our study, NetFlow V5 was used, which contains the following eight fields for each flow after preprocessing: 1) Source IP address, 2) Destination IP address, 3) Source port, 4) Destination port, 5) Protocol, 6) Octets, 7) Packets, and 8) Duration. The IT department at Missouri S&T collects NetFlow data of all users for troubleshooting network connections and policy enforcement. The Missouri S&T campus has a connection to both the standard commodity Internet and the Internet 2 education research network. Both Internet and Internet 2 traffic pass through the same router where NetFlow statistics recording and exporting are enabled. Every five minutes, these flows are exported from the router to a collector where they are stored for a period of 45 days for analysis purposes before being discarded automatically. In order to obtain the NetFlow data of participants, the flows pertaining to each participant were identified based on the source IP field, and subsequently filtered and logged to a secure remote server at the end of every month. As the Missouri S&T campus uses a DHCP (Dynamic Host Configuration Protocol) to provide IP address, the IP address used by a participant at one time could be used by someone else later. Therefore, the extraction process begins by creating a mapping file and associating each user with a set of assigned IP addresses, along with the start and end time stamps. This information is used by a backup daemon to extract user-specific NetFlow information by filtering flows based on the source IP field. The mapping file is created by analyzing DHCP logs that include a
2

Since we do not process IP addresses, websites visited were not accessed. Hence associations between visits to Social Networking sites like Facebook, Twitter etc. and depressive symptoms were not investigated in this study. This is part of future research.

participants user-id, which is that participants campus email address. Note that this process, summarized in Figure 1, was executed by the Missouri S&T campus IT department. This process was completely automated. Subsequently, the Internet usage of each participant indexed by appropriate pseudonyms (as discussed in Section II) was

delivered to the research team. In this study, the Internet data used was the one collected in February 2011, the month in which the depressive symptoms of participants were surveyed.

Figure 1: Illustration of the overall NetFlow data logging process

IV.

INTERNET FEATURES EXTRACTION

The NetFlow data sample for a single participant is shown in Table II. Each row in the table corresponds to a flow.
TABLE II. srcIP

SAMPLE NETFLOW DATA PER PARTICIPANT

dstIP Prot Srcp dstp oct pkts Dur

granularity is low, these features can be used to answer questions like: Does more Internet usage associate with increased depressive symptoms? In our study, aggregate flow statistics were derived using the flow-report in the flow-tools suite. Additionally, bash scripting was used to extract the data and convert it into a feature vector, one per participant. In total, 14 features were derived as summarized in Table III. B. Application Level Features Traffic aggregation alone has low granularity. For example, an aggregate of high email and low chatting usage may appear similar to an aggregate of low email and high chatting usage. Application-level statistics capture more information by sub-categorizing aggregate traffic features by application. In other words, traffic features such as flows, octets, packets and duration are derived per application such as http, email, peer-to-peer (p2p), chat, etc. A total of 61 applications were identified by filtering flows based on set combinations of destination port and destination protocol fields, as allocated by IANA (Internet Assigned Numbers Authority) [17]. Since NetFlow data was only logged for on-campus Internet usage, some application categories like socks, squid, and blubster, showed little or no activity. Universities tend to block such services due to security and copyright issues, and students also tend to limit such activities on campuses. In our study, 25 applications were hence retained. These applications were further grouped into eight categories, as summarized in Table IV.

131.151.x.x 131.151.x.x 131.151.x.x 131.151.x.x

208.78.x.x 208.78.x.x 208.78.x.x 208.78.x.x

6 6 6 6

65055 65058 65042 65062

80 80 80 443

1187 1141 402 1533

13 12 5 9

158 166 67 196

NetFlow data in its natural form is unsuitable for statistical analysis. In order to derive meaningful statistics, we have to preprocess NetFlow data D={flowi}i=1:k for each participant into an Ndimensional feature vector. Also, as the number of rows associated with a participant approaches millions when aggregated over a month, preprocessing also compresses the data into manageable proportions. As the space of all possible feature vectors is large, care must be taken to extract features that are likely to associate with depressive symptoms. We derived three broad features inspired by related research in the Psychological Sciences community (discussed in Section I-A), as discussed below. A.

Aggregate Traffic Features

The simplest feature is a representation of overall aggregate traffic statistics, such as total packets, flows and octets. Although the

TABLE III. Feature

AGGREGATE FEATURES
Description

flows Total Flows oct Total Octets pkts Total Packets timeflows Total Time (1/1000 secs) (flows) durreal Duration of data (realtime) durdata Duration of data (1/1000 secs) aftime Average flow time (1/1000 secs) apsize Average packet size (octets) afsize Average flow size (octets) apflow Average packets per flow afsec Average flows / second (flow) afsecreal Average flows / second (real) akbits Average Kbits / second (flow) akbitsreal Average Kbits / second (real)

plots, while Levenes test was used to assess the equality of variance. If the data deviated from a normal distribution, the non-parametric Mann-Whitney U-test was used. Table V contains Internet usage features that correlate statistically with depressive symptoms (i.e., CES-D score 16). Results from Mann-Whitney U-test are presented next.
TABLE V.

INTERNET USAGE FEATURES THAT STATISTICALLY CORRELATE WITH DEPRESSIVE SYMPTOMS (CES-D SCORES 16)
Kendall tau-b average packets per flow .056 .137* .198* p2p octets .173* .075 .111 p2p packets .236** .106* .160* p2p duration .265** .098 .143 chat octets .267** .100 .145 mail packets .164* .050 .068 mail duration .202** .048 .064 ftp duration .267** .100 .145 remote file octets .281** .117* .172* **Correlation is highly significant at 0.01 level (2-tailed) *Correlation is significant at 0.05 level (2-tailed) *Correlation is significant at 0.05 level (2-tailed) Internet Features Pearson Spearman Rho

TABLE IV. Category

CATEGORIES OF APPLICATION FEATURES

Applications

File-sharing applications based on peer-to-peer p2p architecture (edonkey, neomodus, winmx) HyperText Transfer Protocol applications (http, http https) Stream media applications (shoutcast, real, streaming winmedia, stream-works, audiogalaxy) chat Instant messaging applications (aim, irc, carracho) email Email traffic (IMAP, POP3, SMTP) ftp File transfer applications (snmp, ftp) Massively multiplayer online games (battlenet, gaming quake, starseige, portzero, halflife, gamespyarcade, directx) remote file Remote file system access (afs, nfs) access C. Entropy Based features Difficulty concentrating or making clear decisions is a symptom of depression among college students [27]. We capture randomness in Internet usage via Shannon Entropy (H). Intuitively, entropy estimates the average uncertainty of a series of discrete events. Given a discrete random variable X, Shannon entropy H(X) is:

H(X) =

x P(x) log (P(x)),

(1)

Mann-Whitney U-test revealed statistically significant difference in the mean values of average packets per flow across subjects with and without depressive symptoms (U (216) = 2231, Z = -2.384, (2tailed) = 0.017). Subjects with depressive symptoms have higher average packets per flow ( = 168.47, = 46.11) compared to those without symptoms ( = 110.91, = 14.51). Among Application features, Mann-Whitney U-test revealed statistically significant differences in the mean values of remote file octets across participants with and without depressive symptoms (U (216) = 2343, Z = -1.989, (2-tailed) = 0.047). Participants with depressive symptoms have higher remote file octets ( = 1.17 1010, = 1.88 1010) when compared to those without symptoms ( = 5.90 109, = 5.97 109). Additionally, when applications were directly analyzed (instead of categories), Mann-Whitney U-test revealed significant mean value differences in internet relay chat (irc) octets (U (216) = 2602, Z = -2.225, (2-tailed) = 0.026); packets (U (216) = 2596, Z = -2.269, (2-tailed) = 0.023); and duration (U (216) = 2608, Z = -2.182, (2-tailed) = 0.029). For the entropy based features, Mann-Whitney U-test revealed statistically significant mean value differences for flow duration entropy across subjects with and without depressive symptoms (U (216) = 2337.5, Z = -2.008, (2-tailed) = 0.045). The results are summarized in Table VI.
TABLE VI.

where, P(x) is the probability that X is in state x. In our study, we compute the Entropy for all eight fields in a NetFlow record: 1) Source IP address, 2) Destination IP address, 3) Source port, 4) Destination port, 5) Protocol, 6) Octets, 7) Packets and 8) Duration. V.
RESULTS FROM STATISTICAL ANALYSIS

INTERNET USAGE FEATURES WITH SIGNIFICANT MEAN VALUE DIFFERENCE ACROSS SUBJECTS WITH AND WITHOUT DEPRESSIVE SYMPTOMS FROM MANN-WHITNEY U-TEST
Internet Features U (216) Z (2-tailed)

Statistical analysis was performed to correlate the Internet usage data collected, with CES-D scores (both collected in February 2011). For each feature derived, Pearsons, Spearman Rho, and Kendall taub correlation coefficients were determined. Additionally, T-tests were attempted to identify Internet usage features that significantly differentiated participants that exhibited depressive symptoms from those that did not. The T-test assumes a normal data distribution and homogeneity of variance. Normality was verified by observing P-P

Average packets per flow Remote File Octets IRC (Chat) octets IRC (Chat) packets IRC (Chat) duration Flow Duration Entropy

2231 2343 2602 2596 2608 2337.5

-2.384 -1.989 -2.225 -2.269 -2.182 -2.008

0.017 0.047 0.026 0.023 0.029 0.045

VI.

INTERPRETATIONS AND POSSIBLE APPLICATIONS OF

OUR FINDINGS

B.

Applications of our Findings

There are a number of applications of our methodology and findings from this paper. We present three below. Investigating associations between other mental health disorders and Internet usage: Our methodology is general and can be used to study associations between Internet usage and other mental health disorders like anorexia, bulimia, ADHD, schizophrenia etc. We could also investigate associations between other Internet features like visits to social networking sites, late night Internet use, and randomness in Internet usage times etc. with depressive symptoms. Proactively discovering depressive symptoms from passive and unobtrusive Internet usage monitoring: Using the correlating Internet usage features derived in our study, we are currently attempting to build a classifier to proactively discover depressive symptoms among students by passive, unobtrusive and run-time monitoring of their Internet usage. The classifier can be readily deployed at campus settings to discover depressive symptoms among students. If privacy is a concern, it is straightforward to deploy the classifier into students personal computers also. Privacy is implicit, and the student can voluntarily seek support services. We believe that such an approach will significantly improve both the quality and timeliness of detection and treatment of mental disorders today. Designing Internet (or Computer) based interventions for depression: There are many recent studies exploring Internet based intervention strategies for alleviating depression [28, 29, 30]. Our findings in this paper could yield new insights on designing and administering effective Internet based interventions for depressive disorders. Our work can also enable run-time adaptation of intervention strategies based on severity of symptoms for a subject. Furthermore, our findings will impact the evaluation of Internet based intervention strategies. With our findings, one could easily test the efficacy of Internet based intervention strategies by verifying corresponding changes in correlating Internet usage features identified in this study. This, we believe will positively impact the design of effective Internet based interventions in the future. VII.

A.

Interpretations of our Findings

In this section, we present some practical interpretations to our findings by integrating them with existing research in the Psychological Sciences community on college students depression and Internet usage. Average packets per flow: The average packets per flow is high when a large number of packets are generated per flow. Larger number of packets is typical under Internet streaming and downloading, which is common when watching videos and gaming. This is intuitive, as gaming and video watching are common symptoms of Internet addiction that lead to depressive symptoms [18, 19, 20]. Another interpretation is in terms of application switching behavior. High average packets per flow can occur when there is frequent switching between internet applications. Frequent switching may be related with difficulty concentrating, which is also an indicator of depressive symptoms among students [27]. Peer-to-Peer usage: The correlation observed between peer-to-peer usage and depressive symptoms is intuitive. Sharing files like music, movies, photos etc. are primary reasons for using peer-to-peer services. Students are prone to be addicted to such kinds of content, which may explain this trend. Chatting: Excess online chatting can affect the psychology of young people, and can also cause social isolation and loneliness in the real world, potentially leading to depressive symptoms [21, 22]. People with depression are also known to join Depression Chat Rooms to overcome their feelings of isolation. This may explain Chat octets being significantly high for students with depressive symptoms. Email: Excessive email usage identified in our study as statistically correlating with depressive symptoms is supported by studies in [22]. Frequent email checking may relate with high levels of anxiety, which in-turn correlates with depressive symptoms. It is also theorized that email addiction is a form of impulsive-compulsive disorder in the sense that victims (especially young people) suffer from a compulsive and irresistible need to check messages (often even in the middle of the night). Flow Duration Entropy: As discussed before, difficulty concentrating or making clear decisions are indicators of depressive symptoms among students [27]. When Flow Durations have high entropy, it is likely a result of frequent switching among multiple Internet applications, which is likely to result in highly variable flow durations, and hence high Entropy. Ftp and Remote File usage: It is not completely clear why ftp duration and remote file octets correlate with depressive symptoms. One interpretation could be that since excess ftp usage and remote file octets are indicative of excess file transfers, this could indicate addiction to certain types of files that may associate with depressive symptoms. In our study, we do not access the content of files exchanged, and hence we are limited in the nature of conclusions derived here. Interestingly though, ftp packets and ftp octets did not show statistically significant correlations; only the ftp duration did. Our on-going studies attempt to further explain these trends based on more discussions with counselors, clinical psychologists and educators, and with more experiments with larger subject sizes.

CONCLUSIONS

In this paper, we report findings from a study conducted at Missouri S&T on associating depressive symptoms among college students with their Internet usage. We have identified that average

packets per flow, peer-to-peer (octets, packets and duration),

chat octets, mail (packets and duration), ftp duration, and remote file octets show statistically significant correlations with depressive symptoms. Additionally, Mann-Whitney U-tests revealed that average packets per flow, remote file octets, chat (octets, packets and duration) and flow duration entropy have a statistically significant difference in the mean values across groups with and without depressive symptoms. To the best of our knowledge, this is the first study associating Internet usage with depressive symptoms among college students using real Internet data collected in an unobtrusive and privacy preserving manner. VIII.

ACKNOWLEDGEMENTS

Partial support of the National Science Foundation, the Missouri University of Science & Technology Intelligent Systems Center, and the Mary K. Finley Missouri Endowment is gratefully acknowledged. The authors also thank Mr. Levi Malott for executing a number of scripts for Internet data processing.

REFERENCES
1. American College Health Association. American College Health Association National College Health Assessment (ACHANCHA) Web Summary. Updated August 2007. Available at http://www.acha-ncha.org/data highlights.html 2. http://psychcentral.com/blog/archives/2010/09/02/statisticsabout-college-depression/ 3. A.J. Campbell, S.R. Cumming and I. Hughes. Internet use by the socially fearful: Addiction or therapy? CyberPsychology & Behavior, 9(1):6981, 2006. 4. K.S. Young and R.C. Rogers. The relationship between depression and Internet addiction. CyberPsychology & Behavior, 1(1):2528, 1998. 5. C. Morgan and SR Cotten. The relationship between internet activities and depressive symptoms in a sample of college freshmen. Cyberpsychology & Behavior: The impact of the Internet, multimedia and virtual reality on behavior and society, 6(2):133, 2003. 6. L. Lam and Z. Peng. Effect of Pathological Use of the Internet on Adolescent Mental Health. Archives of Pediatrics & Adolescent Medicine, 2010. 7. C. Morrison and H. Gore. The Relationship between Excessive Internet Use and Depression: A Questionnaire-Based Study of 1319 Young People and Adults. Psychopathology, 2010. 8. S.B. Stevens and T.L. Morris. College dating and social anxiety: Using the Internet as a means of connecting to others. CyberPsychology & Behavior, 2007. 9. M. Griffiths and R.T.A. Wood. Risk factors in adolescence: The case of gambling, videogame playing, and the Internet. Journal of Gambling Studies, 16(2):199-225, 2000. 10. R.A. LaBrie, H.J. Shaffer, D.A. LaPlante, and H. Wechsler. Correlates of college student gambling in the United States. Journal of American College Health, 52(2):53-62, 2003. 11. K. Bessiere. Effects of Internet Use on Health and Depression: A Longitudinal Study. Journal of Med Internet Research, 12(1):e6, 2010. 12. http://www.universityofcalifornia.edu/news/article/22290 13. J.E. Gangwisch, D. Malaspina, K. Posner, L.A. Babiss, S.B. Heymsfield, J.B. Turner, G.K. Zammit, and T.G. Pickering. Insomnia and Sleep Duration as Mediators of the Relationship between Depression and Hypertension Incidence. American Journal of Hypertension, 23(1):62-69, 2009. 14. L.S. Radloff. The CES-D scale: A self report depression scale for research in the general population. Applied psychological measurement, 1(3):385401, 1977. 15. K.L. Bradley, A.L. Bagnell, and C.L. Brannen. Factorial validity of the center for epidemiological studies depression 10 in adolescents. Issues in Mental Health Nursing, 31(6):408412, 2010. 16. R E. Roberts, J.A. Andrews, P.M. Lewinsohn, and H. Hops. Assessment of depression in adolescents using the center for epidemiologic studies depression scale. Psychological Assessment: A Journal of Consulting and Clinical Psychology, 2(2):122, 1990. 17. IANA, Internet Assigned Numbers Authority [online]. Available from: http://www.iana.org/assignments/port-numbers. 18. A. Boals. Correlates of Video Game Addiction. PhD thesis, University of North Texas 2010. 19. J. Kim, R. LaRose, and W. Peng. Loneliness as the cause and the effect of problematic internet use: The relationship between internet use and psychological well-being. CyberPsychology & Behavior, 12(4):451455, 2009.

20. J.B. Weaver III, D. Mays, S. Sargent Weaver, W. Kannenberg, G.L. Hopkins, D. Eroglu, and J.M. Bernhardt. Health-risk correlates of video-game playing among adults. American Journal of Preventive Medicine, 37(4):299305, 2009. 21. L. Bonetti, M.A. Campbell, and L. Gilmore. The relationship of loneliness and social anxiety with childrens and adolescents online communication. CyberPsychology, Behavior, and Social Networking, 13(3):279285, 2010. 22. J. Morahan-Martin and P. Schumacher. Loneliness and social uses of the Internet. Computers in Human Behavior, 19(6):659 671, 2003. 23. http://www.educause.edu/resources/educatingthenetgeneration/co nveniencecommunicationsandco/6070. 24. S Edwards, L Tinning, JSL Brown, Reluctance to seek help and the perception of anxiety and depression in the United Kingdom: a pilot vignette study. The Journal of Nervous and Mental Disease, 2007. 25. D. Joyce and S. Weibelzahl. Text messaging as a means to lowering barriers to help-seeking in students with depression, Proceedings of IADIS International Conference e-Society, Dublin, Ireland, 2006. 26. http://pewinternet.org/reports/2011/college-students-andtechnology/report.aspx 27. http://www.nimh.nih.gov/health/publications/depression-andcollege-students/complete-index.shtml. 28. D.J. Wantland, C.J. Portillo, W.L. Holzemer, et.al. The effectiveness of Web-based vs. non-Web-based interventions: a meta-analysis of behavioral change outcomes. Journal of Medical Internet Research, 2004. 29. G. Anderson and P. Cuijpers. Internet-based and other computerized psychological treatments for adult depression: a meta-analysis. Cognitive Behaviour Therapy, 2009. 30. N. Titov and B. Dear. Trans-diagnostic internet treatment for anxiety and depression: A randomized controlled trial. Behavior Research and Therapy, 2011 31. M.G. Michael and K. Michael. The Fall-Out from Emerging Technologies: on Matters of Surveillance, Social Networks and Suicide. IEEE Technology and Society Magazine 30.3 (2011): 15-18.

AUTHOR BIOS
Raghavendra Kotikalapudi is a Software Development Engineer at Microsoft Corp. His research interests center around Artificial Intelligence and their applications to human-centric computing. He received an MS degree in Computer Science from Missouri S&T. Sriram Chellappan is an Assistant Professor in the Computer Science Dept. at Missouri S&T. His research interests are in Network Security and Social Computing. He is a member of IEEE. Frances Montgomery is a Curators Teaching Professor in the Psychological Sciences Dept. at Missouri S&T. Her research interests are adolescent psychology, maladaptive attitudes and interpersonal problems in teens. Donald Wunsch is the M.K. Finley Missouri Distinguished Professor in Computer Engineering at Missouri S&T. His major research interests are in Clustering, Reinforcement Learning and Fuzzy Systems. He is a Fellow of the IEEE. Karl Lutzen is in the Information Technology Department at Missouri S&T. He is a Certified Information Systems Security Professional (CISSP #87137). He also teaches Undergraduate/ Graduate Classes on Network security, Forensics and Privacy.