Northeast India is characterized by a unique and complex tectonic framework, making it one of the most seismically active regions in the world. This region’s tectonic setting, comprising thrusts such as the Main Frontal Thrust (MFT), Main Boundary Thrust (MBT), Main Central Thrust (MCT), (Gupta et al., 2022; Kayal, 2008) and faults like the Dauki (Kayal et al., 2006), Kopili, and Oldham Faults (England and Bilham, 2015), highlights the necessity for advanced Seismic Hazard Assessment (SHA) techniques. The existing methods, such as the conventional logic tree approach for selecting and weighting Ground Motion Prediction Equations (GMPEs), present significant limitations in addressing the region’s distinctive geological and seismological characteristics.

In recent studies, various methodologies have been applied to assess seismic hazards in Northeast India. Mishra et al. (2024) and Kumar et al. (2023a) conducted a comprehensive seismic hazard assessment for Guwahati and Silchar city, employing both probabilistic and deterministic approaches to estimate ground motion parameters such as peak ground acceleration (PGA) and spectral acceleration at various probabilities of exceedance. Both studies employed equal weightage to each Ground Motion Prediction Equation (GMPE) to estimate ground motion parameters. Lallawmawma et al. (2023) used different areal sources along with multilayer logic tree approach for the computation seismic hazard in Mizoram. Ghione et al. (2021) used a hybrid model to compute hazard for Bhutan and Northeast India considering shallow and subduction zone. Kumar et al. (2023b) employed a probabilistic approach for the evaluation of seismic hazard for all the capital city of Northeast India. Here, also equal weightage has been given to the two used GMPE without considering the tectonic importance of the different study region of Northeast India. Agrawal et al. (2023) added a key factor social vulnerability for the evaluation of risk assessment of Northeast India. In this study, two different GMPE were used for two different focal depths of source zone for the determination of PGA. As factors affecting GMPE are not limited to only focal depth, so the use of GMPE based on focal depth does not lead to accurate measurement of PGA. All the above recent literature puts a serious question on the assigned weights for each of the GMPEs used in various studies.

The traditional logic tree method relies heavily on subjective expert judgment to assign weights to GMPEs, leading to inconsistencies and potential biases. Additionally, it oversimplifies the intricate relationships between key seismic parameters like magnitude, distance, site conditions, and tectonic settings, failing to adequately account for uncertainties (epistemic and aleatory) in peak ground acceleration (PGA) predictions. This gap is particularly critical for Northeast India, where plate interactions at the tri-junction of the Indian, Eurasian, and Burmese plates result in a high degree of crustal deformation and frequent large-magnitude earthquakes along with shallow intra-slab Bengal basin earthquakes. The inability of existing approaches to account for these uncertainties undermines the reliability of seismic hazard models, which are crucial for infrastructure planning and disaster risk mitigation.

The primary objective of this study is to develop and implement a Seismic Hazard Assessment (SHA) framework using Fuzzy Multi-Criteria Decision Making (MCDM) techniques to enhance the accuracy and reliability of Peak Ground Acceleration (PGA) estimates in Northeast India. The study aims to minimize subjectivity in GMPE selection and weighting through a systematic Fuzzy MCDM approach, integrate uncertainties related to site conditions, tectonic settings, magnitude scaling, and distance attenuation, and improve PGA estimation accuracy in the seismotectonically complex region. It tailors the framework to account for the high seismicity of the Indo-Burma Range, Shillong Plateau, Eastern Himalayas, Mishmi Thrust, Naga Thrust, and Bengal Basin while generating robust seismic hazard maps for better infrastructure design, urban planning, and disaster preparedness. Additionally, the study demonstrates the superiority of Fuzzy MCDM over conventional methods in handling uncertainties and providing more reliable hazard estimates, ultimately contributing to a deeper understanding of seismic hazards and improving seismic risk mitigation in Northeast India.

The TrF-FUCOM- Neutrosophic-TOPSIS method is a hybrid MCDM technique used in this work. The TrF-FUCOM approach was first presented by Majumder (2023a). Saaty created the Analytic Hierarchy Process (AHP), a popular MCDM method, earlier in 1980. AHP assigns suitable weights to several criteria using a defined technique and pairwise comparisons. To address this challenge, Saaty extended the AHP into the Analytic Network Process (ANP) in 1996, with a revised version published in 2001. ANP is particularly effective for complex multi-criteria decision-making (MCDM) problems, as it incorporates feedback and interactions both within and between clusters. This method enables a systematic approach to handling interdependencies in decision-making systems, making it a more generalized form of AHP that accounts for intricate relationships among attributes and decision levels. One of the MCDM methods, Simple Additive Weighting (SAW), was proposed by Zionts and Wallenius (1983). The Step-wise Weight Assessment Ratio Analysis (SWARA) method, introduced by Keršuliene et al. (2010), is a relatively recent and efficient approach for determining subjective weights in multi-criteria decision-making (MCDM) problems. Compared to methods like AHP, SWARA offers a lower computational complexity, making it a more straightforward alternative in certain applications. The weights of alternatives are usually determined via n ( n − 1 ) / 2 comparisons. Rezaei introduced the Best-Worst Method (BWM), another MCDM technique, in 2015. Rezaei (2015) claims that BWM overcomes several of AHP’s drawbacks. Because BWM only requires 2 n − 3 comparisons, it is more efficient and consistent than AHP, which requires a greater number of pairwise comparisons. Furthermore, because fewer comparisons are required to produce trustworthy findings, BWM provides more flexibility by getting beyond the restriction of comparing a maximum of nine criteria.

By using vectors like “best-to-others” and “others-to-worst”, the BWM improves reliability and produces more reliable findings than the AHP, which mostly uses pairwise comparisons. However, BWM could have trouble figuring out precise weight coefficient values when large swings affect the consistency level. Interval values should be calculated in these situations, and the average of these intervals should be used to determine the final weight coefficient (Rezaei, 2015). The core region of the interval may or may not contain the required weight coefficients. Interval weight values might not accurately reflect ideal weight coefficients in situations of inconsistency, according research by Pamučar et al. (2018a). Pamučar et al. (2018b) presented the Full Consistency Method (FUCOM), a technique for establishing criteria weights, in order to overcome these issues. FUCOM tackles the shortcomings of the AHP and BWM models. Pairwise comparisons are made easier while maintaining a balanced number of comparisons ( n − 1 ) that are neither too few nor too many. When determining criteria weights, the FUCOM technique incorporates the subjective preferences of the decision-maker, especially in the first steps where criteria are assessed and compared pairwise. FUCOM is effective, however its criteria weight values vary slightly (Pamučar et al., 2018b). By eliminating the requirement for repeated pairwise comparisons of criteria, FUCOM offers a major advantage (Božanic et al., 2019; Božanić et al., 2020). Linguistic variables are frequently chosen over exact numerical values in circumstances when decision-makers lack information or expertise. Zadeh (1965) proposed the idea that linguistic variables offer a mathematical framework for dealing with imperfect information. Under ambiguous or imprecise circumstances, fuzzy sets (FSs) have shown themselves to be a useful tool for representing MCDM problems. Triangular or trapezoidal fuzzy numbers can be used to assess the possible results for each criterion (Guha and Chakraborty, 2011). Trapezoidal fuzzy numbers can also be triangular fuzzy numbers, according to Zheng et al. (2012). A trapezoidal fuzzy number must take on at least one of its extreme values in order to produce a triangular fuzzy number. In the development of methodologies, researchers have found that trapezoidal fuzzy numbers (TrFNs) provide notable advantages over triangular fuzzy numbers (Majumder, 2023a, 2023b; Majumder et al., 2023a). Majumder (2023a) presented the TrF-FUCOM methodology in light of these discoveries.

Hwang and Yoon first proposed OPSIS, or Technique for Order Preference by Similarity to Ideal Solution, in 1981. Yoon then refined the technique in 1987. Hwang et al. made more improvements in 1993. The chosen alternative should have the largest distance from the Negative Ideal Solution (NIS) and the shortest distance from the Positive Ideal Solution (PIS), according to the fundamental tenet of TOPSIS. In the year 2023, Neutrosophic-TOPSIS under Single Valued Neutrosophic Set (SVNS) environment was developed by Pramanik et al. (2023) to ascertain the ranking of alternatives. In order to facilitate efficient determination, the TrF-FUCOM technique directs the process of allocating criteria weights (Majumder, 2023a). TOPSIS is used in this study because it is easy to use and adaptable to both qualitative and quantitative needs. By comparing the best and worst performance of each alternative, this method improves the trustworthiness of ranking results. TOPSIS is a great option for situations where the interaction between cost and performance is crucial since it also takes cost-benefit parameters into account.

The Seismic Hazard Assessment (SHA) framework utilizing Fuzzy Multi-Criteria Decision Making (MCDM) techniques is highly relevant to the seismotectonic settings of Northeast India and offers significant novelty and advantages in addressing the region’s complex seismic hazard challenges. Northeast India is one of the most seismically active regions in the world due to its unique tectonic setting, involving the interaction of the Indian Plate with the Eurasian and Burma Plates, resulting in high seismicity across the Indo-Burma Range (IBR), Shillong Plateau (SP), Eastern Himalayas (EH), Mishmi Thrust (MT), Naga Thrust (NT), and Bengal Basin (BB). The conventional Deterministic Seismic Hazard Assessment (DSHA) process often struggles to account for the inherent uncertainties in seismic parameters, such as site conditions, tectonic settings, magnitude scaling, and distance attenuation, which are critical for accurate PGA estimation.

The Fuzzy MCDM framework addresses these challenges by incorporating fuzzy sets and membership functions to quantify uncertainties and reduce subjectivity in weight assignments for Ground Motion Prediction Equations (GMPEs). This approach is particularly advantageous for Northeast India, where the seismotectonic complexity and variability in fault systems demand a more nuanced and flexible hazard assessment methodology. By integrating multiple criteria and fuzzy logic, the framework provides a more reliable and realistic estimation of PGA, capturing the uncertainties associated with seismic events and tectonic settings. This is especially important for regions like the Indo-Burma Range and Mishmi Thrust, where high PGA values are observed due to intense tectonic activity, and the Shillong Plateau, where the Dauki Fault contributes to significant seismic hazard.

In this study, two different fuzzy-based MCDM techniques are used to determine the score values of the considered alternatives. The initial preparation is divided into two parts: the preliminary analysis of Trapezoidal Fuzzy Numbers (TrFN) and the preliminary analysis of Single-Valued Neutrosophic Sets (SVNS), which are discussed in Sections 2.1.1 and 2.1.2, respectively.

This study aims to develop and implement a Seismic Hazard Assessment (SHA) framework using Fuzzy MCDM techniques to improve the accuracy and reliability of PGA estimates. The new PGA equation is developed as the average weighted sum of the considered PGA values. The PGA is determined by the formula (1) with the help of weights of considering PGA and its calculated PGA from ground motion prediction equation (GMPE) The weights are determined with the help of a hybrid MCDM technique. Figure 1 illustrates the computational steps of PGA. (1) PGA f = ∑ i = 1 m p i × v ( PGA GMPE − i ) ∑ i = 1 m p i , where: PGA f = Final PGA; p i = Calculated Weight for ith Ground motion Prediction Equation; v ( PGA GMPE − i ) = calculated PGA from ith No of Ground motion Prediction Equation ( i = 1 ( 1 ) m ).

To determine the weights of each GMPE ( PGA GMPE − i ), in the present study we use two MCDM techniques, namely Trapezoidal Fuzzy Full Consistency Method (TrF-FUCOM) and Neutrosophic-TOPSIS under Single Valued Neutrosophic Set (SVNS) environment (Neutrosophic-TOPSIS). The TrF-FUCOM is used to determine the weights of criteria and Neutrosophic-TOPIS Strategy within the SVNN used to determine the weights of alternatives. In the Model-I and Model-II, we discuss the computational process of TrF-FUCOM and Neutrosophic-TOPIS Strategy respectively.

Model-I: In this model, we discuss how to determine the weights of criteria with the help of TrF-FUCOM.

Phase-i: Suppose F = { ξ r : r = 1 ( 1 ) n } is a collection of criteria.

Phase-ii: Decision-makers’ (DMs’) preferences and opinions about the importance of each criterion are used to prioritize them in the first phase. The top rank is given to the factor with the highest weight coefficient, and the lowest weight coefficient factor is ranked last. The most significant factors are listed first, followed by the least significant, in descending order of their weight coefficients. As a result, the qualities are ranked as ξ r ( 1 ) ≻ ξ r ( 2 ) ≻ ⋯ ≻ ξ r ( κ ) , and κ represents the rank of a certain factor. An equality symbol in place of ‘≻’ is used to show that two or more components have the same rank.

Phase-iii: By using Table 1, it is possible to compare factors according to their rankings, which are mostly established by the component that is rated first. Determining the trapezoidal fuzzy criterion importance ( ℘ ˜ ξ r ( κ ) ) for each factor is made easier by this procedure. The other components are compared a total of n − 1 times because the most significant element is self-compared (its importance is ℘ ˜ ξ r ( 1 ) = E I). Based on the criteria’s indicated importance, Eq. (2) is used to compute the fuzzy comparative significance χ ˜ κ / ( κ + 1 ) . (2) χ ˜ κ / ( κ + 1 ) = ℘ ˜ ξ r ( κ + 1 ) ℘ ˜ ξ r ( κ ) = ( ℘ ˜ ξ r ( κ + 1 ) a , ℘ ˜ ξ r ( κ + 1 ) b , ℘ ˜ ξ r ( κ + 1 ) c , ℘ ˜ ξ r ( κ + 1 ) d ) ( ℘ ˜ ξ r ( κ ) a , ℘ ˜ ξ r ( κ ) b , ℘ ˜ ξ r ( κ ) c , ℘ ˜ ξ r ( κ ) d ) , where the significance of the factor ranked ξ r ( κ ) with respect to the factor ranked ξ r ( κ + 1 ) is shown by the symbol χ ˜ κ / ( κ + 1 ) .

A Trapezoidal Fuzzy vector that represents the relative relevance of choice factors is thus produced by the expression (3). (3) χ ˜ = ( χ ˜ 1 / 2 , χ ˜ 2 / 3 , … , χ ˜ κ / ( κ + 1 ) ) .

Phase-iv: The Trapezoidal Fuzzy weight coefficients of components ( y ˜ 1 , y ˜ 2 , … , y ˜ n ) T were finally calculated. However, the final values of the weight coefficients must satisfy the following conditions:

Only when the transitivity between the weight coefficients is fully satisfied does the minimum DMC (deviations from the maximum consistency), i.e. η = 0, get satisfied. Therefore, it can be concluded that y ˜ κ y ˜ κ + 1 − χ ˜ κ / ( κ + 1 ) = 0, as well as y ˜ κ y ˜ κ + 2 − χ ˜ κ / ( κ + 1 ) ⊗ χ ˜ ( κ + 1 ) / ( κ + 2 ) = 0. The DMC for these weight coefficient values is η = 0. Finding the values of the weight coefficients of criteria ( y ˜ 1 , y ˜ 2 , … , y ˜ n ) T that meet the requirement that | y ˜ κ y ˜ κ + 1 − χ ˜ κ / ( κ + 1 ) | ⩽ η and | y ˜ κ y ˜ κ + 2 − χ ˜ κ / ( κ + 1 ) ⊗ χ ˜ ( κ + 1 ) / ( κ + 2 ) | ⩽ η with the minimization of value η are necessary in order to meet these requirements.

The final nonlinear model for identifying the best fuzzy values for the assessment criteria’s weight coefficients can be set to ( y ˜ 1 , y ˜ 2 , … , y ˜ n ) T based on the parameters specified. (6) Min η s.t. | y ˜ κ y ˜ κ + 1 − χ ˜ κ / ( κ + 1 ) | ⩽ η , for all r = 1 ( 1 ) n , | y ˜ κ y ˜ κ + 2 − χ ˜ κ / ( κ + 1 ) ⊗ χ ˜ ( κ + 1 ) / ( κ + 2 ) | ⩽ η , for all r = 1 ( 1 ) n , ∑ r = 1 n y ˜ r = 1 , 0 ⩽ y r a ⩽ y r b ⩽ y r c ⩽ y r d , for all r = 1 ( 1 ) n , where y ˜ r = ( y r t 1 , y r t 2 , y r t 3 , y r t 4 ) and χ ˜ κ / ( κ + 1 ) = ( χ κ / ( κ + 1 ) t 1 , χ κ / ( κ + 1 ) t 2 , χ κ / ( κ + 1 ) t 3 , χ κ / ( κ + 1 ) t 4 ).

The requirement that y ˜ κ y ˜ κ + 1 − χ ˜ κ / ( κ + 1 ) = 0, as well as y ˜ κ y ˜ κ + 2 − χ ˜ κ / ( κ + 1 ) ⊗ χ ˜ ( κ + 1 ) / ( κ + 2 ) = 0 meet must be met in order to attain the highest consistency. Thus, a fuzzy linear model, Eq. (7), can be created from the model provided by Eq. (6). If it is solved, the ideal fuzzy values of the weight coefficients are ( y ˜ 1 , y ˜ 2 , … , y ˜ n ) T . (7) Min η s.t. | y ˜ κ − y ˜ κ + 1 ⊗ χ ˜ κ / ( κ + 1 ) | ⩽ η , for all r = 1 ( 1 ) n , | y ˜ κ − y ˜ κ + 2 ⊗ χ ˜ κ / ( κ + 1 ) ⊗ χ ˜ ( κ + 1 ) / ( κ + 2 ) | ⩽ η , for all r = 1 ( 1 ) n , ∑ r = 1 n y ˜ r = 1 , 0 ⩽ y r a ⩽ y r b ⩽ y r c ⩽ y r d , for all r = 1 ( 1 ) n , where y ˜ r = ( y r t 1 , y r t 2 , y r t 3 , y r t 4 ) and χ ˜ κ / ( κ + 1 ) = ( χ κ / ( κ + 1 ) t 1 , χ κ / ( κ + 1 ) t 2 , χ κ / ( κ + 1 ) t 3 , χ κ / ( κ + 1 ) t 4 ).

Transform optimal weights ( y ˜ ∗ 1 , y ˜ ∗ 2 , … , y ˜ ∗ n ) T in which y ˜ ∗ r = ( y r t 1 ∗ , y r t 2 ∗ , y r t 3 ∗ , y r t 4 ∗ ) (for all r = 1 ( 1 ) n) into crisp value by making use of Eq. (8). (8) R ( y ˜ ∗ r ) = y r t 1 ∗ + 2 y r t 2 ∗ + 2 y r t 3 ∗ + y r t 4 ∗ 6 , for all r = 1 ( 1 ) n .

Model-II: In this model, we discuss how to determine the weights of an alternative with the help of Neutrosophic-TOPSIS.

Consider the set of alternatives U = { u i : i = 1 ( 1 ) m }, i ⩾ 1 and F = { ξ r : r = 1 ( 1 ) n }, r ⩾ 2 is the set of attributes with weights R ( y ˜ ∗ r ), r = 1 ( 1 ) n respectively.

Decision-makers assign ratings to the u i , i = 1 ( 1 ) m alternatives based on the attributes ξ r , r = 1 ( 1 ) n, which are represented using an SVNN. Let’s assume the rating for the r t h attribute concerning the i t h alternative is presented as follows:

u i r ∗ = ( ξ r , T u i ( ξ r ) , I u i ( ξ r ) , F u i ( ξ r ) ), i = 1 ( 1 ) m, r = 1 ( 1 ) n, where 0 ⩽ T u i ( ξ r ) + I u i ( ξ r ) + F u i ( ξ r ) ⩽ 3. Here, ( T i r , I i r , F i r ) is denoted as an SVNN u i r ∗ ,( i = 1 ( 1 ) m and r = 1 ( 1 ) n), where r represents the number of attributes and i represents the number of alternatives. Derive the decision matrix based on the ratings: Ω ∗ = [ u i r ∗ ] m × n

The TOPSIS approach has been summed up as outlined below:

Step 1: The score-matrix Ω = [ u i r ] m × n ,( i = 1 ( 1 ) m and r = 1 ( 1 ) n) is obtained from the decision matrix Ω ∗ = [ u i r ∗ ] m × n utilizing the methods outlined in the preceding section: i.e. u i r = S 1 ( u i r ∗ ).

Step 2: Evaluation of the normalized decision matrix T = [ t i r ] m × n using Eq. (9). (9) t i r = u i r ∑ r = 1 n u i r , i = 1 ( 1 ) m and r = 1 ( 1 ) n .

Step 3: Figure out the θ = [ λ i r ] m × n weighted normalized decision matrix, where λ i r = R ( y ˜ ∗ r ) ⊗ t i r , i = 1 ( 1 ) m and r = 1 ( 1 ) n.

Step 4: Find the Neutrosophic Positive Ideal Solution (NPIS) and Neutrosophic Negative Ideal Solution (NNIS), which are degraded by ∂ + , as well as ∂ − , respectively. ∂ + = { λ 1 + , λ 2 + , … , λ n + } , where λ r + = max r λ i r , r = 1 ( 1 ) n and ∂ − = { λ 1 − , λ 2 − , … , λ n − } , where λ r − = min r λ i r , r = 1 ( 1 ) n .

Step 5: Use the following formulas (10) and (11) to determine the distance between each NPIS and NNIS option: (10) ρ i + = ∑ r = 1 n ( λ i r − λ i + ) 2 , i = 1 ( 1 ) m , (11) ρ i − = ∑ r = 1 n ( λ i r − λ i − ) 2 , i = 1 ( 1 ) m .

Step 6: Use formula (12) to calculate each alternative’s performance score: (12) δ i = ρ i − / ( ρ i + + ρ i − ) , i = 1 ( 1 ) m .

Step 7: Based on the performance scores, rank the options; they should be in ascending order. Eq. (13) normalizes the performance score. The normalized performance score is used as a final weighting factor for several GMPE equations in this study. (13) p i = δ i ∑ i = 1 m δ i , i = 1 ( 1 ) m . Putting the value of p i in Eq. (1) to enhance the accuracy and reliability of Peak Ground Acceleration (PGA) estimates.

To study the northeastern region of India, several researchers have categorized this area into multiple zones based on its seismic history, seismicity patterns, and the inter- and intra-dynamic movement geometry. This study, however, classifies Northeast India and its surrounding areas into Six primary tectonic domains: the Eastern Himalayan source zone, the Mishmi source zone, Naga thrust zone, the Shillong Plateau source zone, the Bengal basin zone and the Indo-Burmese source zone (Fig. 2). Understanding the tectonic framework of Assam and its neighbouring states necessitates the development of a comprehensive seismotectonic map, grounded in geological field surveys.

The Geological Survey of India initially produced a regional geology map for the entire northeastern region and adjacent areas, utilizing satellite imagery. For detailed seismotectonic analysis of NE India, morphotectonic lineaments across the entire Northeast region were mapped and correlated with SEISAT, 2000 data. Figure 3, created using the MapInfo software, illustrates all major faults. It becomes evident from this figure that three prominent tectonic belts converge at two distinct boundaries—Himalayan and Indo-Burmese—that intersect at the Assam Syntaxis. The region within latitudes 20°N–31°N and longitudes 86°E–98°E encompasses two major mountain ranges: the Eastern Himalayas in the north and the Indo-Burmese range in the southeast. Figures 2 and 3 depict the Eastern Himalayan collision zone, trending east–west to east-northeast–west-southwest, north of latitude 27°N. This zone includes the south-dipping Main Central Thrust (MCT) and Main Boundary Thrust (MBT). Current tectonic compression in this area is approximately north–south, as indicated by studies conducted by Heidbach et al. (2005, 2007), Bilham and Gaur (2000), and Jade et al. (2007), which confirmed the north–south crustal shortening.

South of latitude 26°N and east of longitude 92°E lies the Indo-Burmese subduction zone, where the Indian Plate is descending steeply beneath the Burmese arc, as described by Mukhopadhyay and Dasgupta (1988). To the south of Assam, the Arakan Yoma Belt trends north-northwest to south-southeast, while the Naga Hills lie to the northeast with a northeast–southwest orientation. This tectonic framework reveals interactions between the western Burmese arc, the Sunda plate, and the Burmese plate, as noted by Gahalaut and Gahalaut (2007). Variations in thrusting and right-lateral slip along the western Burmese arc’s north–south trending structures are also significant, as highlighted by Kayal et al. (2004).

Near 27°N and 96°E, in the Assam Syntaxis region, the MBT, Naga Thrust, and the Belt of Schuppen run parallel on either side of the Upper Assam Valley. Figures 2 and 3 show that Assam is surrounded by multiple thrust zones: the MBT and MCT in the north, the Lohit-Mishmi Thrust in the northeast, the Arakan Yoma Belt in the southeast, and the Naga, Disang, and Eastern Boundary Thrusts in the east. Within this region, the intraplate domain is deformed, particularly in the Shillong Plateau and Mikir Hills (24°N–26°N) and in the Tripura Fold Belt (91°E–94°E). This zone is highly seismically active, except for the Assam Gap, as noted by Khattri (1993) and Kayal et al. (2006). The Tripura-Mizoram Fold Belt is defined by its north–south trending anticlines and synclines, shaped by tectonic compression forces. Its seismic activity is significantly influenced by its closeness to active fault systems, including the Dauki Fault near the Shillong Plateau. This fault, along with others such as the Kaladhan Fault, Madhupur Fault (located in Bangladesh), and the Sylhet Fault in Bangladesh, contribute to the seismic risks for Tripura. The Dauki Fault of Meghalaya and the wrench-fold structure identified along the Chittagong coastline, referred to as the Chittagong Coastal Fault (as discussed by Maurin and Rangin, 2009, and Sikder and Alam, 2003), add to the region’s tectonic complexity. Additionally, faults along Tripura’s northern boundary, near the India–Bangladesh border, further highlight its geologically active nature.

Notably, Tripura lies near both the Sylhet Fault and the Chittagong Coastal Fault, which have historically been associated with significant seismic events. The region has witnessed two major earthquakes with magnitudes exceeding Mw 7 and another event measuring Mw 6.9 in the past. This proximity to multiple active fault systems underscores the seismotectonic complexity and elevated seismic risk in the area.

The earthquake catalog used in this study includes detailed information on each event, such as its location (in terms of coordinates), date, time, and magnitudes expressed in various scales (ML, Mw, Ms, and Mb). To assess regional seismicity parameters and other related seismological metrics, a substantial database is essential. The dataset utilized here is derived from Sil et al. (2013), which covers data until 2010, and is supplemented with updated information until 2015, sourced from agencies such as USGS, IMD, ISC, and NGRI. Figure 3 illustrates earthquakes with magnitude Mw greater than 6 that have occurred in Northeast India (NEI). The catalog encompasses a 500 km radius from Assam, covering latitudes 18°–29° N and longitudes 86°–97° E. A total of 6663 earthquake events, recorded from 1761 to 2015, were included in the dataset. Some of these events were initially recorded in different magnitude scales (Mw and others). To standardize the dataset, a conversion to the moment magnitude scale (Mw) was performed using the correlation formula developed by Sitharam and Sil (2014). After transforming the magnitudes to a unified Mw scale, aftershocks and foreshocks were excluded based on the method outlined by Gardner and Knopoff (1974), resulting in a selection of 2882 main shocks with magnitudes greater than Mw 4. According to van Stiphout et al. (2012), the Gardner and Knopoff method follows a Poisson distribution and uses both time and space windows to remove aftershocks and foreshocks. The conversion equation used in this study, derived from historical seismic data specific to the Northeast India region, is the most suitable for converting magnitudes to Mw for the analysis conducted here.

Ground motion models are designed with a regional perspective, considering distinct tectonic features such as subduction zones, intraplate settings, and plate boundary interactions. These models rely on accelerogram recordings collected from various distances and magnitudes across the study region. A regression analysis is then conducted on peak ground acceleration (PGA) values, incorporating both magnitude and distance parameters to minimize variability and derive region-specific coefficients for different spectral periods. Based on seismic activity and tectonic characteristics, the study area is categorized into two primary zones: shallow crustal intraplate/interplate regions and subduction-related in-slab/interface regions. Accordingly, separate attenuation models are necessary to represent ground motion behaviour accurately in each category.

Several researchers have developed ground motion attenuation relationships specific to different regions of India. Bajaj and Anbazhagan (2019) introduced an equation tailored to the Himalayan region, calibrated for earthquakes ranging from magnitude 4 to 9 and rupture distances spanning 10–750 km. This model was formulated using an extensive dataset of 4,940 recorded seismic events, facilitating a comprehensive assessment of regional attenuation characteristics. Likewise, Nath et al. (2012) derived a ground motion model for Northeast India, utilizing data from 13 earthquakes with magnitudes between Mw 4.8 and Mw 8.1, recorded at distances extending up to 100 km. Atkinson and Boore (2003) formulated a GMPE for stable continental regions in eastern North America, which, despite its original focus, can be adapted for the Bengal Basin due to its comparable tectonic stability. Raghu Kanth and Iyengar (2007) proposed a GMPE specifically for Peninsular India, estimating seismic ground motion parameters such as PGA. While developed for a different setting, its application to the Bengal Basin is feasible due to tectonic similarities. Additionally, Singh et al. (2016) devised a GMPE for Northeast India, with a focus on seismically active regions influenced by the Indo-Burmese Arc. Given the tectonic similarities between Tripura and the surrounding active zones, this GMPE is considered applicable to the study area, though local geological variations, including soil properties and site conditions, must be accounted for to ensure accurate ground motion predictions.

To validate the five selected attenuation models (Atkinson and Boore, 2003; Raghu Kanth and Iyengar, 2007; Nath et al., 2012; Singh et al., 2016; Bajaj and Anbazhagan, 2019), strong motion data was sourced from the Cosmos Virtual Data Centre (http://db.cosmos-eq.org/). Seismic recordings from seven significant earthquakes occurring between 1987 and 1997, captured through the SMART-1 digital network covering the Indo-Burma range and Northeast India, were used for this validation.

Seismic recordings obtained from multiple monitoring stations were utilized to assess the accuracy of the selected ground motion prediction equations (GMPEs). A comparative analysis was performed using peak ground acceleration (PGA) values recorded at the rock level for the India-Bangladesh Border earthquake (1988-02-06, 14:50:45 UTC, Mw = 6), juxtaposing them with the predictions from the chosen GMPEs (Fig. 5). However, due to the proximity between the observed data and the modelled predictions (Fig. 5), along with inherent uncertainties in each equation, selecting the most suitable model presents a challenge. These epistemic uncertainties can be mitigated through the logic tree framework, which enables the incorporation of multiple models into the evaluation process.

A logic tree structure consists of branches and nodes, where each branch corresponds to a distinct model, and several branches merge at a common node. To refine the decision-making process, an enhanced hybrid trapezoidal neutrosophic-based approach, referred to as the Integrated TrF-FUCOM- Neutrosophic-TOPSIS method, is applied. This technique assigns weightage to each model branch based on its associated uncertainties, ensuring a more robust and objective evaluation. The incorporation of this advanced methodology aids in addressing the uncertainties linked to seismic hazard assessments. Further details regarding the GMPE equations utilized in this study are provided in Table 4. Figure 6 depicts the decision hierarchy in this case.

The objective of this study is to develop an improved Seismic Hazard Assessment of Northeast India framework by integrating a fuzzy MCDM technique. From the case study it is clear that in the present study consider six sites, namely Bengal Basin, Indo-Burma, Shillong Plateau, Eastern Himalaya, Mishmi Thrust and Naga Thrust. The behaviour considering criteria and alternatives are changed as per Seismotectonic setting. So, in the present study developed 12 models to determine the score value of alternatives. In this study { Site condition ( ξ 1 ) , Tectonic setting ( ξ 2 ) , Magnitude scaling ( ξ 3 ) , Distance Atten- uation ( ξ 4 ) } consider as a set of criteria and { GMPE − 4 ( u 1 ) , GMPE − 3 ( u 3 ) , GMPE − 4 ( u 1 ) , GMPE − 1 ( u 4 ) , GMPE − 5 ( u 5 ) } is consider as a set of alternative.

Model-I: Determine the weights of { ξ 1 , ξ 2 , ξ 3 , ξ 4 } for the location Bengal Basin using TrF-FUCOM.

Model-II: Determine the weights of { ξ 1 , ξ 2 , ξ 3 , ξ 4 } for the location Indo-Burma using TrF-FUCOM.

Model-III: Determine the weights of { ξ 1 , ξ 2 , ξ 3 , ξ 4 } for the location Shillong Plateau using TrF-FUCOM.

Model-IV: Determine the weights of { ξ 1 , ξ 2 , ξ 3 , ξ 4 } for the location Eastern Himalaya using TrF-FUCOM.

Model-V: Determine the weights of { ξ 1 , ξ 2 , ξ 3 , ξ 4 } for the location Mishmi Thrust using TrF-FUCOM.

Model-VI: Determine the weights of { ξ 1 , ξ 2 , ξ 3 , ξ 4 } for the location Naga Thrust using TrF-FUCOM.

Model-VII: Determine the score value of { u 1 , u 2 , u 3 , u 4 , u 5 , u 6 } for the location Bengal Basin using Model I and Neutrosophic-TOPSIS.

Model-VIII: Determine the score value of { u 1 , u 2 , u 3 , u 4 , u 5 , u 6 } for the location Indo-Burma using Model II and Neutrosophic-TOPSIS.

Model-IX: Determine the score value of { u 1 , u 2 , u 3 , u 4 , u 5 , u 6 } for the location Shillong Plateau using Model III and Neutrosophic-TOPSIS.

Model-X: Determine the score value of { u 1 , u 2 , u 3 , u 4 , u 5 , u 6 } for the location Eastern Himalaya using Model IV and Neutrosophic-TOPSIS.

Model-XI: Determine the score value of { u 1 , u 2 , u 3 , u 4 , u 5 , u 6 } for the location Mishmi Thrust using Model V and Neutrosophic-TOPSIS.

Model-XII: Determine the score value of { u 1 , u 2 , u 3 , u 4 , u 5 , u 6 } for the location Naga Thrust using Model VI and Neutrosophic-TOPSIS.