Inter national J our nal of Inf ormatics and Communication T echnology (IJ-ICT) V ol. 15, No. 4, December 2026, pp. 1726 ∼ 1735 ISSN: 2252-8776, DOI: 10.11591/ijict.v15i4.pp1726-1735 ❒ 1726 An explainable AI-based intrusion detection system f or cyber thr eats in stock mark ets using the SECURE framew ork Anandhi Ramani 1 , Malath y S. 1 , Ramya K. 2 1 PG and Research Department of Computer Science, Dw araka Doss Go v erdhan Doss V aishna v Colle ge, Chennai, India 2 PG Department of Information T echnology and BCA, Dw araka Doss Go v erdhan Doss V aishna v Colle ge, Chennai, India Article Inf o Article history: Recei v ed Feb 2, 2026 Re vised Sep 6, 2026 Accepted Sep 18, 2026 K eyw ords: Articial intelligence Cybersecurity Explainable AI Intrusion detection system Mark et manipulation ABSTRA CT As electronic trading technology accelerates the w ay trades occur in nancial mark ets, the ability to manipulate the mark ets in man y dif ferent w ays has be- come stronger through this same technology . T w o w ays this happens are via spoong and layering, which perv ert the w ay prices are determined, confuse in- v estors, and reduce the le v el of trust in the inte grity of nancial mark ets. Exist- ing surv eillance mechanisms were predominantly rule-bas ed or relied on opaque machine learning models, limi ting their adaptability and interpretability . This study proposed SECURE, an e xplainable articial intelligenc e (AI)-based in- trusion detection system designed to detect and interpret spoong and layering acti vities in real time. The frame w ork inte grated ensemble machine learning models, c ybersecurity-inspired monitoring logic, and shaple y additi v e e xpla- nations (SHAP)-based e xplainability to achie v e accurate detection with trans- parent decision-making. Using Le v el-II order book data, the system analysed beha vioural and microstructural trading features to classify manipulati v e pat- terns while generating feature-le v el e xplanations for each alert. Experimental e v aluation demonstrated that the propose d frame w ork achie v ed high detection accurac y , lo w f alse-positi v e rat es, and dependable real-time performance. The results indicated that combining intrusion detection principles with e xplainable AI impro v ed re gulatory interpretability and strengthened the ef fecti v eness of au- tomated mark et surv eillance systems. This is an open access article under the CC BY -SA license . Corresponding A uthor: Anandhi Ramani PG and Research Department of Computer Science, Dw araka Doss Go v erdhan Doss V aishna v Colle ge Chennai, T amil Nadu, India Email: sekadhi@gmail.com 1. INTR ODUCTION In the li st of these c ybersecurity threats, some methods of mark et manipulation lik e spoong and layering pose a signicant threat to mark et inte grity and in v estor condence [1]. In case of spoong, lar ge b uy and sell orders are placed b ut there is no intention behind them for actual e x ecution of trade while in case of layering, percei v ed [2] depth of the mark et is manipulated through placing v arious orders at v arying price le v els [3]. Such manipulati v e techniques distort the process of price disco v ery and create articial liquidity within the nancial mark ets. Retail in v estors can be easily manipulated through these strate gies since the y mainly use publicly a v ailable order book data for their trades [4]. J ournal homepage: http://ijict.iaescor e .com Evaluation Warning : The document was created with Spire.PDF for Python.
Int J Inf & Commun T echnol ISSN: 2252-8776 ❒ 1727 Ho we v er , the current mark et surv eillance technologies hea vily use rule-based and threshold-based ap- proaches for recognizing suspicious beha viors in the mark et [5]. While such an approach is quite successful when detecting specic violati ons, it f ails to e v olv e in response to the ne w and increasingly comple x manipula- tions in the nancial mark et. Additionally , using strict rule-based criteria often leads to a lar ge amount of f alse positi v es, causing an increased w orkload for the re gulators and, thus, decreasing the ef cienc y of automatic mark et surv eillance technologies. Therefore, it w as decided to use articial intelligence (AI) and machine learning algorithms to perform mark et surv eillance. F or dealing with the interpretability issues associated with traditional AI algorithms, the methods of e xplainable articial intelligence (XAI) ha v e recei v ed considerable interes t in the domain of nancial surv eil- lance [6]. One such method is SHAP , which e xplains the contrib ution of indi vidual features to w ards the model prediction [7]. Despite the adv antages of XAI for making the model more interpretable, pre vious research does not consider t h e implementation of e xplainability along with real-time surv eillance and c yber security based intrusion detection system [8]. Dri v en by suc h research issues, this paper a ims to introduce SECURE, an intrusion detection frame- w ork based on c ybersecurity concepts for spoong and layering in stock mark ets. Xu and Lin [9] This frame- w ork ut ilizes machine learning, structured v alidation with rul es, and XAI for real-time and accurate detection and monitoring in stock mark ets [10]. By applying beha vioral analysis together with e xplainable decision sup- port tools, SECURE intends to impro v e transparenc y in the mark et, increase re gulatory control, and of fer better protection for in v estors. The ne xt section describes past research studies and their limitations [11]. T able 1 presents a comparati v e summary of representati v e studies in mark et manipulation dete ction. T able 1 sho ws that signicant achie v ements ha v e been reached in terms of detection of mark et manipulations using rule-based surv eillance, machine learning approaches, c ybersecurity-based intrusion detection system techniques, and e xplainable AI methods. Although man y adv ancements ha v e been made in this eld, there is still a lack of research in terms of creating a holistic surv eillance frame w ork instead of just focusing on enhancing certain components of the detection frame w ork. The machine learning approach usually performs well in terms of detection b ut lacks interpretability while the e xplanation-based methods hardly e v er pro vide real-time operation capability . On the other hand, c ybersecurity-based intrusion detecti on methods ha v e not been used much for beha vioural analysis of Le v el-II order books data. T able 1. Comparison of e xisting studies on mark et manipulation detection Citations F ocus K e y strength K e y limitation [12], [13] Mark et v olatility , trading acti vity during c yber -attacks Sho ws ho w c yber threats impact nancial mark ets Does not detect manipulation patterns [14], [15] Order book depth, spoong patterns, f ak e liquidity Strong theoretical basis for spoong and layering Not real-time operational [16], [17] Cancellation b ursts, order lifetime, frequenc y patterns Detects high-frequenc y spoong strate gies Limited interpretability [18], [19] Layering depth, order placement imbalance Strong re gulatory rele v ance for manipulation cases High computational o v erhead [20], [21] Distrib ution shifts in order -o w No v el quantitati v e method for spoong detection Comple x to tune, not industry-ready [22], [23] HFT order inno v ations, manipulation beha viours Clear distinction between le gitimate HFT and ab use No ML or XAI inte gration [24], [25] System vulnerabilities, attack v ectors in nance Strong justication for IDS inclusion No trading-pattern detection [26], [27] Sector -le v el c ybersecurity challenges Supports c ybersecurity-moti v ated mark et surv eillance Not focused on trading microstructure [28], [29] High-frequenc y trade sequences Strong ML foundation for real-time fraud detection Black-box, lacks e xplainability [30], [31] Price-v olume anomalies, manipulation signals Modern ML approach for manipulation detection Does not handle spoong patterns e xplicitly Se v eral research limitations are identied from the e xisting literature re vie w concerning the problem of stock mark et manipulation detection. The rst one is that the traditional rule-based systems ha v e transpar - ent decision process b ut lack e xibility in adapting to changing spoong and layering techniques in dynamic mark ets [21], [22]. Secondly , despite the f act that machine learning and deep learning models ha v e consider - ably enhanced detection accurac y , the majority of these models remain black-box systems and thus can hardly An e xplainable AI-based intrusion detection system for cyber thr eats in stoc k mark ets ... (Anandhi Ramani) Evaluation Warning : The document was created with Spire.PDF for Python.
1728 ❒ ISSN: 2252-8776 be used for conducting in v estig ation and audi ting. Thirdly , current XAI research is predominantly aimed at pro viding e xplanation after making the predictions and it is seldom incorporated in the mark et surv eillance process. Lastly , while IDS ha v e been pro v ed to be successful in c ybersecurity applications, their application for analysing the beha viour of Le v el-II order book and traders has been barely e xplored. 2. THE PR OPOSED SECURE FRAMEW ORK 2.1. Ov er view of the pr oposed SECURE framew ork In this paper , we introduce the SECURE frame w ork which is a c ybersecurity-inspired and an e xplain- able approach to detect spoong and layering manipulations in the stock mark et as sho wn in Figure 1. Jacob- Leal and Hanaki [23] in contrast to traditional approaches for mark et surv eillance, which use either rule-based monitoring only or machine learning alone, SECURE utilizes ensemble machine learning techniques along with beha vioral rule v alidation and state-based intrusion detection, all under one detection frame w ork, together with XAI techniques. SECURE architecture is made up of six interacting layers, namely the surv eillance layer , the e xamination layer , the control layer , the understanding layer , the response layer , and the enhancement layer . Figure 1. Layers of the SECURE Frame w ork 2.2. Operational w orko w of the SECURE framew ork The proposed SECURE frame w ork follo wed a structured six-stage operational pipeline, as ill u s trated in Figure 2: Surv eillance: real-time ingestion of Le v el-II order book e v ents; Examination: feature computation and preliminary anomaly scoring; Control: rule-based enforcement using predened beha vioural thresholds; Understanding: generation of SHAP-based e xplanations; Response: dispatch of structured and e xplainable alerts; Enhancement: continuous learning through v alidated analyst feedback. Figure 2. Operational w orko w of the SECURE frame w ork Int J Inf & Commun T echnol, V ol. 15, No. 4, December 2026: 1726-1735 Evaluation Warning : The document was created with Spire.PDF for Python.
Int J Inf & Commun T echnol ISSN: 2252-8776 ❒ 1729 2.3. Explainability mechanism XAI w as implemented using SHAP to generate feature-le v el attrib utions [29]. The follo wing is ho w the e xplanation function w orks: f ( x t ) = ϕ 0 + n X i =1 ϕ i (1) where ϕ i represents the contrib ution of the i -th feature to the nal model prediction. This formulation enabled transparent interpretation of ho w indi vidual features inuenced the model output. 2.4. Pr oposed detection models Three progressi v ely adv anced model architectures were implemented. 2.4.1. Model 1: Random f or est with SHAP (Baseline) This baseline model applies supervised learning combined with post-hoc e xplainability [30]. b y t = f R F ( x t ) (2) 2.4.2. Model 2: SECURE + Random f or est + SHAP This model inte grates rule-based e v aluation with anomaly scoring [31]. The operational process is described as follo ws: R t = g ( x t ) (3) A t = | x t − µ | σ (4) S t = α f RF ( x t ) + β A t + γ R t (5) S t ≥ τ ⇒ Sus picious (6) 2.4.3. Model 3: SECURE + Random f or est + SHAP + IDS The nal model, algorithm 1 represents the inte gration of a state-a w are intrusion detection system for session-le v el beha vioural cor -relation [32]. Algorithm 1 IDS-Inte grated SECURE Detection Requir e: Real-time stream S , IDS engine, RF model Ensur e: Alerts and audit logs 1: Initialise IDS session memory 2: f or each e v ent x t do 3: Update session state 4: Extract features 5: Compute R t and A t 6: Predict RF condence 7: Compute IDS correlation score 8: Compute Risk t 9: if Risk t e xceeds thresholds then 10: Generate SHAP e xplanation 11: Create alert and store logs 12: end if 13: end f or 14: Periodically retrain using v alidated feedback C t = IDS ( x t , H t ) (7) Risk t = w 1 P t + w 2 A t + w 3 C t + w 4 R t (8) Risk t ≥ τ high ⇒ Critical Alert (9) Risk t ≥ τ med ⇒ W arning Alert (10) An e xplainable AI-based intrusion detection system for cyber thr eats in stoc k mark ets ... (Anandhi Ramani) Evaluation Warning : The document was created with Spire.PDF for Python.
1730 ❒ ISSN: 2252-8776 3. METHOD 3.1. Dataset description A Le v el-II stock mark et order book dataset w as emplo yed in this study . Each transaction record contained a timestamp, trading symbol, order price, quantity , order type, e x ecution status, order book depth le v el, a unique order identier , v olume at the price le v el, and a deri v ed risk score. The e xperimental data set contains Le v el-II stock mark et order book data with around 800,000 trading instances collected through elec- tronic trading. In this data set, le gitimate trading instances are mix ed with labelled instances of manipulation, which include spoong and layering actions. All transacti on s in the data set ha v e been characterized by mark et microstructure properties, beha vioural measures, and risk properties. The dataset comprised both le gitimate trading instances and labelled e v ents reecting spoong and layering beha viours [24]. Malicious labels were assigned based on e xpert-dened beha vioural rules deri v ed from prior literature to ensure consistenc y with realistic mark et dynamics. 3.2. Dataset a v ailability and r epr oducibility Due to re gulatory , le g al, and condentiality constraints associated with Le v el-II stock mark et data, the ra w order book dataset used in this study cannot be publicly released. T o address reproducibility concerns, the e xperimental dataset emplo yed in this study consists of a combination of anon ymised real trading records and synthetically generated manipulation samples. The synthetic spoong and layering e v ents were generated using e xpert-dened beha vioural rules deri v ed from established nancial microstructure literature, ensuring that realistic mark et dynamics and temporal characteristics were preserv ed. All preprocessing steps, feature engineering procedures, model architectures, and e v aluation protocols are fully described within this paper to enable methodological reproducibility . Furthermore, the proposed SECURE frame w ork is model-agnostic and can be applied to an y Le v el-II order book dataset with equi v alent feature structure. 3.3. Data pr epr ocessing Ra w trading data were processed to ensure consistenc y , reliability and suitability for model de v el- opment [25]. The follo wing preprocessing steps were applied: remo v al of duplicate and logically inconsistent records, missing v alue imputation using statistical substitution, Min–Max normalisation of numerical attrib utes, label encoding of cate gorical v ariables, and temporal ordering of transactions to preserv e sequence inte grity . F ormally , each ra w record x t w as transformed using a preprocessing function as follo ws: x ∗ t = P ( x t ) (11) where x ∗ t represents the cleaned and normalised data instance. 3.4. F eatur e engineering After preprocessing, structured features were deri v ed to capture microstructural and beha vioural char - acteristics associated with spoong and layering [26]. Feature e xtraction w as performed o v er both short-term and session-based time windo ws [27]. The feature v ector w as dened as follo ws: x t = [ q t , p t , l t , c t , b t , v t , f t , r t , ∆ p t ] (12) where q t denotes the order size, p t represents the price relati v e to the mid-price, l t corresponds to the order lifetime, c t indicates the cancellation rate, b t denotes the order book imbalance, v t represents the traded v olume at the price le v el, f t indicates the trader submission frequenc y , r t represents the risk score, and ∆ p t denotes the short-term price mo v ement. 3.5. Machine lear ning detection engine This detection engine uses ensemble tree classier techniques such as Random F orest and Gradient Boosted T rees due to their resilience to high dimensional structured nancial data, as well as SHAP based e xplainability capability . This technique w as trained by the feature v ector creat ed from Le v el-II order book data for classication of trading beha vior into eithe r le gitimate, spoong, or layering beha vior [28]. The predicti v e function is formalised as: b y t = f RF ( x t ) (13) where b y t ∈ { le gitimate , spoong , layering } . Int J Inf & Commun T echnol, V ol. 15, No. 4, December 2026: 1726-1735 Evaluation Warning : The document was created with Spire.PDF for Python.
Int J Inf & Commun T echnol ISSN: 2252-8776 ❒ 1731 4. RESUL TS AND DISCUSSION The proposed SECURE frame w ork w as e v aluated to assess its detection performance and operat ional feasibility under realistic t rading conditions [33]. Three progressi v ely adv anced model congurations were implemented: Model 1, Model 2, and Model 3. All models were e v aluated using stratied v e-fold cross- v alidation. Performance w as measured using accurac y , precision, recall, F1-score, R OC–A UC, f alse positi v e rate, detection latenc y , and memory usage [34]. 4.1. Quantitati v e perf ormance analysis The T able 2 illustrates e v aluation of all models using stratied v e-fold cross-v alidation. Perform ance w as measured using accurac y , precision, recall, F1-score, R OC–A UC, f alse positi v e rate, detection latenc y and memory usage [34]. T able 2. Performance comparison of model congurations Metric Model 1 Model 2 Model 3 Accurac y (%) 93.1 95.9 97.0 Precision (%) 78.3 85.1 89.4 Recall (%) 70.0 82.0 87.6 F1-score (%) 73.1 83.6 88.5 R OC–A UC (%) 89.3 93.9 96.2 F alse positi v e rate 0.15 0.10 0.06 Latenc y (ms) 5.4 3.2 1.7 Memory usage (MB) ∼ 100 ∼ 140 ∼ 210 Real-time capability No P artial Full 4.2. P erf ormance inter pr etation T able 2 vi vidly represents the performance comparison of model congurations whereas the Figure 3 represents the e v aluati on of models based on their performance. From an operational perspecti v e, these impro v ements are particularly important: higher precision reduces unnecessary re gulatory w orkload, higher recall reduces the risk of undetected manipulation, and lo wer f alse positi v e rates impro v e condence in automated surv eillance. Figure 3. Ev aluation of models based on f alse positi v e rate, latenc y , R OC-A UC, and memory usage An e xplainable AI-based intrusion detection system for cyber thr eats in stoc k mark ets ... (Anandhi Ramani) Evaluation Warning : The document was created with Spire.PDF for Python.
1732 ❒ ISSN: 2252-8776 4.3. Explainability analysis using SHAP 4.3.1. Global featur e importance An e xplainability analysis w as carried out to determine if the prediction beha vior of the frame w ork aligns with what mark et manipulation e xperts kno w about it [35]. Across all model congurations, SHAP anal- ysis consis tently identied the most inuential beha vioural indicators: short-windo w cancellation rate, relati v e order size, order book imbalance surv eillance, trader submission frequenc y , and short-term price mo v ement. The strong alignment between SHAP-deri v ed feature rankings and established nancial microstructure [36] theory conrms that the model’ s internal reasoning reects e xpert kno wledge, which is essential for re gulatory acceptance [37]. 4.3.2. Local trade-le v el explanations At the indi vidual trade le v el, SHAP e xplanat ions indicated that suspicious classications were primar - ily dri v en by: rapid order cancellations, abnormally lar ge order placements, and high-frequenc y submissions. 4.4. Operational system perf ormance The full SECURE model generated real-time alerts in under tw o milliseconds, meeting the opera- tional requirements of e xchange-le v el monitoring [38]. The IDS layer substantially impro v ed alert quality by correlating e v ent s across tim e windo ws and trader identities, thereby reducing redundant alerts and increasing o v erall precision. Arin [39] the feedback-dri v en enhancement module demonstrated adapti v e learning capa- bility . The introduction of a small proportion of analyst-v al idated feedback resulted in: impro v ed prediction stability , reduced f alse alert rates, and f aster con v er gence during retraining. 4.5. Comparati v e analysis among peer system In contrast to traditional rule-bas ed surv eillance systems, which are rigid and prone to alert saturation, SECURE achie v es a balanced inte gration of: adaptability through machine learning, control through rule-based logic, and transparenc y through SHAP-based e xplainability . 4.6. Use-case v alidation T w o representati v e mark et scenarios were used to v alidate practical applicability: a spoong cluster scenario, in which rapid lar ge-order cancellations were accurately detected and clearly e xplai ned [39], and a v olatile mark et layering scenario, where articial order book depth w as successfully identied through IDS- based session correlation [40], [41]. These cases conrm that the frame w ork performs reliably not only in controlled e xperimental settings b ut also under realistic mark et conditions. 5. CONCLUSION SECURE is the name of the XAI-based intrusion detection system proposed in this paper for detecting and e xplaining instances of fraudulent beha viors such as spoong and layering in high-frequenc y trading - nancial mark et en vironment. The proposed methodology combines the po wer of machine learning algorithms, c ybersecurity-based techniques, and XAI to o v ercome the dra wbacks of tradi tional rule-based and opaque sys- tems. Experiments conducted sho w that SECURE pro vides e xcellent detection accurac y , lo w f alse positi v e detection rate, and good computational ef cienc y which allo ws the frame w ork to be used in the real-time mark et surv eillance system. In addition, the use of SHAP-based e xplanations helps mak e the results more interpretable by nding the top important feature contrib utions in each detection. Due to the implementation of an adapti v e feedback loop system, the proposed scheme is able to consist ently update its detection approach to counteract ne wly de v eloped manipulati v e approaches, thus increasing its stability o v er time. Future w ork will in v olv e e xpanding the scope of SECURE frame w ork to include more types of mark et manipulations, such as w ash trading, front running, and quote stuf ng. Future impro v ements in the frame w ork will also in v olv e inte gration of multimodal e xternal data sources lik e nancial ne ws and social media sentiments that will enrich conte xt intelligence. Future w ork will also in v olv e de v eloping the frame w ork with adv ersarial rob ustness, in- creasing computational ef cienc y to scale up to a bigger size and also using real-w orld datasets in association with re gulatory authorities and stock e xchanges to v alidate the proposed frame w ork. Int J Inf & Commun T echnol, V ol. 15, No. 4, December 2026: 1726-1735 Evaluation Warning : The document was created with Spire.PDF for Python.
Int J Inf & Commun T echnol ISSN: 2252-8776 ❒ 1733 FUNDING INFORMA TION The authors ackno wledge the nancial support pro vided by Dw araka Doss Go v erdhan Doss V aishna v Colle ge, Chennai, through the Research De v elopment Cell and colle ge management, which sanctioned seed funding to support the research acti vities reported in this study . No e xternal funding agenc y w as in v olv ed in this w ork. A UTHOR CONTRIB UTIONS ST A TEMENT This journal uses the Contrib utor Roles T axonomy (CRediT) to recognize indi vidual author contrib utions, reduce authorship disputes, and f acilitate collaboration. Name of A uthor C M So V a F o I R D O E V i Su P Fu Anandhi Ramani ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ Malath y S. ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ Ramya K. ✓ ✓ ✓ ✓ ✓ ✓ ✓ C : C onceptualization I : I n v estig ation V i : V i sualization M : M ethodology R : R esources Su : Su pervision So : So ftw are D : D ata Curation P : P roject Administration V a : V a lidation O : Writing - O riginal Draft Fu : Fu nding Acquisition F o : F o rmal Analysis E : Writing - Re vie w & E diting CONFLICT OF INTEREST ST A TEMENT Authors state no conict of interest. D A T A A V AILABILITY The data that support the ndings of this study are a v ailable upon reasonable request. REFERENCES [1] ´ A. Cartea, S. Jaimung al, and Y . W ang, “Spoong and price manipulation in order -dri v en mark ets, ” Applie d Mathematical Finance , v ol. 27, no. 1–2, pp. 67–98, Mar . 2020, doi: 10.1080/1350486X.2020.1726783. [2] G. Mark, “Spoo ng and layering, ” J Corp L , v ol. 45, no. 2, pp. 399–469, 2020. [3] J. D. Montgomery , “Spoong, mark et manipulation, and the limit-order book, ” in SSRN Electronic Journal , 2016, doi: 10.2139/ssrn.2780579. [4] B. L. Do and T . J. Putnins, “Detecting layering and spoong in mark ets, ” in SSRN Electronic Journal , 2023, pp. 1–58, doi: 10.2139/ssrn.4525036. [5] J. T allbo ys, Y . Zhu, and S. Rajase g arar , “Identication of stock mark et manipulation with deep learning, ” in Lecture Notes in Computer Science (including subseries Lecture Notes in Articial Intelligence and Lecture Notes in Bioinformatics) , v ol. 13087 LN AI, 2022, pp. 408–420, doi: 10.1007/978-3-030-95405-5 29. [6] S. Khodabandehlou and S. A. H. Golpaye g ani, “Mark et manipulation detection: a systematic literature re vie w , ” Expert Systems with Applications , v ol. 210, p. 118330, Dec. 2022, doi: 10.1016/j.esw a.2022.118330. [7] Y . Hao, B. V and, B. M. Delg ado, and S. Baldi, “Mark et manipulation in stock and po wer mark ets: a study of indicator -based monitoring and re gulatory challenges, ” Ener gies , v ol. 16, no. 4, p. 1894, Feb . 2023, doi: 10.3390/en16041894. [8] L. N. Mintarya, J. N. M. Halim, C. Angie, S. Achmad, and A. K urnia w an, “Machine lea rning approaches in stock mark et prediction: a systematic literature re vie w , ” Procedia Computer Science , v ol. 216, no. 4, pp. 96–102, 2023, doi: 10.1016/j.procs.2022.12.115. [9] J. Xu and J. Lin, “GDet: le v eraging graph neural netw orks for intelligent detection of trade-based mark et manipulation, ” in Proceedings of the 2024 4th International Conference on Big Data, Art icial Intelligence and Risk Management , Ne w Y ork, NY , USA: A CM, No v . 2024, pp. 716–723, doi: 10.1145/3718751.3718866. [10] A. Mashrur , W . Luo, N. A. Zaidi, and A. Robles-K elly , “Machine learning for nancial risk management: A surv e y , ” IEEE Access , v ol. 8, pp. 203203–203223, 2020, doi: 10.1109/A CCESS.2020.3036322. [11] O. K. T osun, “Cyber -attacks and stock mark et acti vity , ” International Re vie w of Financial Analysis , v ol. 76, p. 101795, Jul. 2021, doi: 10.1016/j.irf a.2021.101795. [12] A. Stenfors, M. Doraghi, C. So vian y , M. Susai, and K. V akili, “Cross-mark et spoong, ” Journal of International Financial Mark ets, Institutions and Mone y , v ol. 83, p. 101735, Mar . 2023, doi: 10.1016/j.intn.2023.101735. [13] X. T ao, A. Day , L. Ling, and S. Drapeau, “On detecting spoong strate gies in high-frequenc y trading, ” Quantitati v e Finance , v ol. 22, no. 8, pp. 1405–1425, Aug. 2022, doi: 10.1080/14697688.2022.2059390. An e xplainable AI-based intrusion detection system for cyber thr eats in stoc k mark ets ... (Anandhi Ramani) Evaluation Warning : The document was created with Spire.PDF for Python.
1734 ❒ ISSN: 2252-8776 [14] P . W eber , K. V . Carl, and O. Hinz, “ Applications of e xplainable articial intelligence in nance—a systematic re vie w of nance, information systems, and computer science li terature, ” Management Re vie w Quarterly , v ol. 74, no. 2, pp. 867–907, Jun. 2024, doi: 10.1007/s11301-023-00320-0. [15] T . T ao, C. Zhang, and G. Zhou, “Spoong detection using w asserstein dis tance, ” Journal of Financial Mark ets , v ol. 54, no. 100554, 2020, doi: 10.1016/j.nmar .2020.100554. [16] V . Dalk o and M. H. W ang, “High-frequenc y trading: order -based inno v ation or manipulation?, ” Journal of Banking Re gulation , v ol. 21, no. 4, pp. 289–298, Dec. 2020, doi: 10.1057/s41261-019-00115-y . [17] R . Hasan, M. A. Hoque, and M. M. Rahman, “Cybersecurity threats in nancial systems, ” Computers & Security , v ol. 91, no. 101706, 2020, doi: 10.1016/j.cose.2020.101706. [18] F . A. Alzahrani and N. A. E. Alyoubi, “Cybersecurity challenges in the nancial sector , ” Journal of King Saud Uni v ersity – Computer and Information Sciences , v ol. 34, no. 8, pp. 5381–5392, 2022. [19] S. A. Adetoun, “Deep learning vs. real-time nancial fraud detection in high-frequenc y trading, ” Journal of Science, T echnology and Engineering Research, v ol. 2, no. 4, pp. 28–40, 2024. [20] Q. Liu, C. W ang, P . Zhang, and K. Zheng, “Det ecting stock mark et manipulation via machine learning: Evidence from China securities re gulatory commission punishment cases, ” International Re vie w of Financial Analysis , v ol. 78, p. 101887, No v . 2021, doi: 10.1016/j.irf a.2021.101887. [21] M. E. V erhulst and J. M. E. Pennings, “Spoong in US futures mark ets: an interdisciplinary approach, ” Capital Mark ets La w Journal , v ol. 20, no. 3, Jul. 2025, doi: 10.1093/cmlj/kmaf012. [22] C. Desagre, F . Laly , and M. Petitjean, “Re visiting the trading acti vity of high-frequenc y trading rms around ultra-f ast ash e v ents, ” Financial Inno v ation , v ol. 11, no. 1, p. 68, Jan. 2025, doi: 10.1186/s40854-024-00726-z. [23] S. Jacob-Leal and N. Hanaki, “ Algorithmic trading, what if it is just an illusion? Evidence from e xperimental asset mark ets, ” Journal of Beha vioral and Experimental Economics , v ol. 112, p. 102240, Oct. 2024, doi: 10.1016/j.socec.2024.102240. [24] A. Indria w an and N. A. Rakhma w ati, “Detecting Manipulation in NFT mark et using graph-based deep learning, ” ECTI T rans- actions on Computer and Information T echnology (ECTI-CIT) , v ol. 19, no. 4, pp. 707–719, Oct. 2025, doi: 10.37936/ecti- cit.2025194.259882. [25] A. Alf ajeer , A. Alta weel, A. Bouridane, D. Mouheb, and S. Aslam, “On detecting stock price manipulation attacks: a comprehensi v e systematic literature re vie w , ” Multimedia T ools and Applications , v ol. 84, no. 29, pp. 35793–35870, 2025, doi: 10.1007/s11042- 025-20641-4. [26] K. Lin and Y . Gao, “Model interpretability of nancial fraud detection by group SHAP , ” Expert Systems with Applications , v ol. 210, p. 118354, Dec. 2022, doi: 10.1016/j.esw a.2022.118354. [27] Y . Zhou, H. Li, Z. Xiao, and J. Qiu, “ A user -centered e xplainable articial intelligence approach for nancial fraud detection, ” Finance Research Letters , v ol. 58, p. 103479, 2023, doi: 10.1016/j.frl.2023.104309. [28] D. V ijayanand and G. S. Smrith y , “Explainable AI - enhanced ensemble learning for nancial fraud detection in mobile mone y transactions, ” Intelligent Decision T echnologies , v ol. 19, no. 1, pp. 52–67, Jan. 2025, doi: 10.1177/18724981241289751. [29] A. Adadi and M. Berrada, “Peeking inside the black-box: a surv e y on e xplainable articial Intelligence (XAI), ” IEEE Access , v ol. 6, pp. 52138–52160, 2018, doi: 10.1109/A CCESS.2018.2870052. [30] J. Jin and Y . Zhang, “The analysis of fraud detection in nancial mark et under machine learning, ” Scientic Reports , v ol. 15, no. 1, p. 29959, Aug. 2025, doi: 10.1038/s41598-025-15783-2. [31] T . A w osika, R. M. Shukla, and B. Pranggono, “T ransparenc y and pri v ac y: the role of e xplainable AI and federated learning in nancial fraud detection, ” IEEE Access , v ol. 12, pp. 64551–64560, 2024, doi: 10.1109/A CCESS.2024.3394528. [32] S. K. Aljunaid, S. J. Almheiri, H. Da w ood, and M. A. Khan, “Secure and transparent banking: e xplainable AI-dri v en federated learning model for nancial fraud detection, ” Journal of Ris k and Financial Management , v ol. 18, no. 4, p. 179, Mar . 2025, doi: 10.3390/jrfm18040179. [33] S. Latha et al. , “ An XAI-po wered approach for nancial fraud detection using anomaly detection and classication techniques, ” Journal of Informati on Systems Engineering and Management , v ol. 2025, no. 55s, pp. 2468–4376, 2024, [Online]. A v ailable: https://www .jisem-journal.com/ [34] M. Muksalmina, A. Syah yana, F . Hidayatullah, G. M. Idroes, and T . R. No viandy , “C redit card fraud detection through e xplainable articial intelligence for managerial o v ersight, ” Indatu Journal of Management and Accounting , v ol. 3, no. 1, pp. 17–28, Jun. 2025, doi: 10.60084/ijma.v3i1.301. [35] X. Q. Chen, C. Q. Ma, Y . S. Ren, Y . T . Lei, N. Q. A. Huynh, and S. Narayan, “Explainable articial intelligence in nance: a bibliometric re vie w , ” Finance Research Letters , v ol. 56, p. 104145, Sep. 2023, doi: 10.1016/j.frl.2023.104145. [36] A. A. Almazroi and N. A yub, “Online payment fraud detection model using machine learning techniques, ” IEEE Access , v ol. 11, pp. 137188–137203, 2023, doi: 10.1109/A CCESS.2023.3339226. [37] J. ˇ Cerne vi ˇ cien ˙ e and A. Kaba ˇ sinskas, “Explainable articial intelligence (XAI) in nance: a systematic literature re vie w , ” Articial Intelligence Re vie w , v ol. 57, no. 8, p. 216, Jul. 2024, doi: 10.1007/s10462-024-10854-8. [38] P . Dini, A. Elhanashi, A. Be gni, S. Saponara, Q. Zheng, and K. Gasmi, “Ov ervie w on Intrusion detection systems design e xploiting machine learning for netw orking c ybersecurit y , ” Applied Sciences , v ol. 13, no. 13, p. 7507, Jun. 2023, doi: 10.3390/app13137507. [39] M. M. Arin, “OLGBM: optuna-optimized light gradient boosting machine for intrusion detection, ” in Proc. IC4ME2, IEEE , 2021. [40] M. A. T alukder et al. , “Machine learning-based net w ork intrusion detection for big and imbalanced data, ” Journal of Big Data , v ol. 11, p. 33, 2024, [Online]. A v ailable: https://doi.or g/10.1186/s40537-024-00886-w [41] D. Dalg ade, N. P atil, M. Joshi, and D. Hiran, “Enhanci ng WSN intrusion detection: tw o-tier feature selection and optuna- opti- mized ensemble learning, ” International Journal of W ireless & Mobile Netw orks , v ol. 17, no. 1, May 2025, doi: 10.21203/rs.3.rs- 6726618/v1. Int J Inf & Commun T echnol, V ol. 15, No. 4, December 2026: 1726-1735 Evaluation Warning : The document was created with Spire.PDF for Python.
Int J Inf & Commun T echnol ISSN: 2252-8776 ❒ 1735 BIOGRAPHIES OF A UTHORS Dr . Anandhi Ramani is an academician in the eld of computer applications with a strong focus on teaching, curriculum de v elopment, and research-oriented learning. She has been acti v ely in v olv ed i n designing outcome-based curricula, mentoring students, and inte grating emer ging tech- nologies such as cloud computing, serv erless computing, c ybersecurity , and data-dri v en applications into academic programs. Her academic interests include impro ving pedagogical practices for di v erse learners, guiding stude nt projects, and contrib uting to research publications and quality assurance acti vities in higher education. She can be contacted at email: sekadhi@gmail.com. Malath y S. is a full-time Ph.D. research scholar in the PG and Research Department of Computer Science at Dw araka Doss Go v erdhan Doss V aishna v Colle ge, Chennai, T amil Nadu, India. She holds a master’ s de gree in Computer Applications and has qualied the UGC-NET e xamination in Computer Science and Applications . Her research interests include c ybersecurity , stock mark et manipulation detection, e xplainable articial intelligence, intrusion detection systems, and nancial data analytics. She can be contacted at email: malath y10112001@gmail.com. Ramya K. is a f aculty member in the eld of computer science with a s trong focus on soft- w are engineering and system modeling. Her research interests include Unied Modeling Language, softw are design methodologies, and the int e gration of emer ging technologies in education and or g a- nizational sys tems. She acti v ely eng ages in scholarly research, academic writing, and conference- oriented studies to enhance both theoretical and practical kno wledge. She participates in f aculty de v elopment and induction programs to strengthen her research and pedagogical competencies. Her academic goal is to contrib ute high-quality research that supports inno v ation, ef fecti v e teaching, and kno wledge dissemination. She can be contacted at email: ramya.dgvc2024@gmail.com. An e xplainable AI-based intrusion detection system for cyber thr eats in stoc k mark ets ... (Anandhi Ramani) Evaluation Warning : The document was created with Spire.PDF for Python.