Highway Proneness Appraisal to Landslides along Taiping to Ipoh Segment Malaysia, Using MCDM and GIS Techniques

Abstract

Landslides are geological hazards that claim lives and affect socio-economic growth. Despite increased slope failure, some constructions, such as road constructions, are still being performed without proper investigation of the susceptibility of slope mass movement. This study researches the susceptibility of landslides in a study area encompassing a major highway that extends from Taiping to Ipoh, Malaysia. After a comprehensive literature review, 10 landslide conditioning factors were considered for this study. As novel research in this study area, multi-criteria decision-making (MCDM) models such as AHP and fuzzy AHP were used to rank the conditioning factors before generating the final landslide susceptibility mapping using Geographical Information System (GIS) software. The landslide susceptibility map has five classes ranging from very low (9.20%) and (32.97%), low (18.09%) and (25.60%), moderate (24.46%) and (21.36%), high (27.57%) and (13.26%), to very high (20.68%) and (6.81%) susceptibility for the FAHP and AHP models, respectively. It was recorded that the area is mainly covered with moderate to very high landslide risk, which requires proper intervention, especially for subsequent construction or renovation processes. The highway was overlayed on the susceptibility map, which concludes that the highway was constructed on a terrain susceptible to slope instability. Therefore, decision-makers should consider further investigation and landslide susceptibility mapping before construction.

1. Introduction

Landslides are types of geological disasters that occur in mountainous terrains on different spatio-temporal ranges that threaten human lives, cause economic loss, and lead to environmental degradation [1,2]. Some studies revealed that landslide dangers are becoming more common over time, with recent developments related to increasing population and climate changes [3]. Landslides are natural disasters that occur due to gravity impact, causing mass movement of rocks, debris, and soil on a downward slope [4]. Recognizing the process and zonation of landslide susceptibility areas is critical in disaster management. It can be used as a common tool to aid the decision-making process.

The geologic environment influences the strength of potential rupture surfaces in rocks and soil, and the probability, process, and classification of earth movements or hazards that may occur are all influenced by the geologic environment [5]. The underlying rocks impact the soils that have evolved in these areas and their susceptibility to disasters [6]. Over 85% of the rocks in West Malaysia are granitic (Figure 1a) and affected by granitoid intrusion and faulting, [7], especially during heavy rainfall, which weakens over time and is susceptible to landslides.

Landslides can be induced by lithology, changes in slope geometry, water level, rainfall intensity, and force, and more. In Malaysia, however, landslides are caused primarily by extreme rainfall [8] due to experiencing high amounts of rainfall yearly because of the country’s tropical location. Different rocks absorb precipitation in various ways. Increasing the soil’s moisture and converting a forested region to a crop-growing area could cause landslides [9,10]. Landslide risk can also be increased by building a road that cuts off the toe of a steep slope. Many landslides have been documented in recent years along Projeck Lebuhraya Utara-Selatan Berhad (PLUS) highways, roads, and streams, scouring the edges of streams [11].

Therefore, landslide susceptibility, hazard, and risk mapping should be implemented to limit landslide occurrence. Landslide susceptibility mapping (LSM) is an effective method for predicting and identifying landslides. Considering that LSM offers valuable insights, decision-makers can implement the significant result in master planning [1]. Furthermore, LSM depicts the places that have the likelihood of experiencing a landslide during a certain time frame or under specific predisposing factors [12]. As a result, thorough landslide prediction along highways mitigates damage by providing preventive measures. These mitigation steps could include early warning and preparation, designating extremely sensitive landslide zones as the red zone for construction purposes, and assessing and investigating landslide events.

Remote sensing and geographical information system (GIS) can be used to map landslides. GIS plays an integral purpose in LSM and enables multiple spatial geoscientific data to be viewed, digitized, and analyzed. However, according to Tella and Balogun [13] and Chakhar and Martel [14], relying solely on GIS-based applications has limitations in the decision-making process, such as the inability to model the relative importance of various competing criteria. Furthermore, Chakhar and Martel [14] stated that GIS does not support decision-makers’ preferences and cannot compare various criteria. Therefore, the solution is to integrate various tools and models with GIS. The data quality is an essential factor to be considered for LSM; nevertheless, determining the type of methods to use is a bit challenging [2]. Since there is no commonly accepted precise methodology or strategy for assessing and predicting landslide hazard patterns, different methods have been used to date for LSM prediction [15].

The methods utilized for LSM can be categorized into qualitative and quantitative approaches. Qualitative procedures are subjective because they solely rely on the expert’s judgment while conducting the susceptibility or hazard assessment [16]. Conversely, quantitative approaches generate a numerical prediction, indicating the probabilities of landslides occurring in a given area [17]. Additionally, LSM has two types of mapping: (i) direct mapping and (ii) indirect mapping. According to Kayastha et al. [18], direct mapping approaches identify the spatial heterogeneity of slope instabilities directly from historical landslides. In contrast, indirect mapping methods predict areas of potential instability using predisposing variables pertinent to landslides.

Qualitative methods such as multi-criteria decision-making (MCDM) models coupled with subjective judgement from experts and decision-makers have been used for LSM. A common approach is the Analytic Hierarchy Process (AHP), which has proven its efficacy in LSM. Despite the uncertainties and biases attributed to AHP [19,20], it has created insights toward mitigating landslides and is being used for LSM [1,21,22,23,24,25,26,27]. However, it is noteworthy that some researchers fuzzified the AHP pairwise comparison for more accurate weight and biases’ sidelining. The fuzzy set theory, which imitates human judgement, was created to eliminate inconsistency and uncertainty in decisions attributed to AHP [28,29,30,31,32]. Several fuzzy-based methods have been devised and utilized in various disciplines recently; an instance is a successful implementation of the Fuzzy Analytical Hierarchical Process (FAHP) method in LSM.

Based on the foregoing, one of the goals of this study is to propose a qualitative model capable of resolving the uncertainties and imprecision of AHP via the representation of decision-makers’ judgment as fuzzy numbers or sets. Therefore, this study compares the efficacy of AHP and FAHP models in landslide susceptibility mapping in the study area. It is noteworthy that despite the wide utilization of AHP and FAHP for LSM, there is a gap in the literature in the study area. Furthermore, it is noteworthy that although some landslide events were recorded along PLUS highways [5,11], there is scarce literature on the sustainability approach. Currently, the highway is in good condition and connects to other states. However, several areas require immediate attention. Furthermore, the slope stability in these areas is of utmost importance, as slight instability could put commuters at risk. As a result, determining the landslide risk along the highway is critical for risk preparedness (that is, pre-landslide events).

This is early research regarding LSM in the study area. This study considers different factors influencing landslides for the LSM to facilitate better mitigation strategies against landslide hazards. Therefore, this study can be useful in managing and preventing landslide events along highways. The findings of this study will help researchers and construction professionals select low-susceptibility areas for construction projects that may be exposed to landslide hazards. In a smart city, landslide hazard maps are used to plan safe and cost-effective transportation infrastructure routes. Thus, this study used AHP and fuzzy AHP for LSM along Taiping to Ipoh highway in Malaysia.

2. Study Area

The North–South Expressway Project, with over 770 km distance, connects numerous important cities and towns, supporting western Peninsular Malaysia’s economic development [33]. The study area encompasses the environs of a section of the North–South Highway that runs from Taiping at 4°51′ N and 101°43′ E of latitude and longitude to Ipoh at 4°35′ N and 101°04′ E of latitude and longitude as illustrated in Figure 1b,c. The distance is approximately 73 km. Malaysia’s climate is characterized by two monsoons, namely: northeast and southwest monsoons. During December, January, and February, the northeast monsoon cycle is Malaysia’s wettest season and the period with the highest flooding. Meanwhile, the southwest monsoon occurs between May and September, when the land is driest, resulting in droughts [8]. Due to the location of this tropical country in the equatorial zone, the mean temperature across the year is generally high (>25 °C), and the humidity is also relatively high due to the hot temperature. Moreover, Malaysia also receives a lot of rain, with over 2400 mm every year [34].

From the sands and silts of the coastal plains to the granite of the Main Collection, Malaysia contains a diverse number of geological structures. Geologists classified the rocks based on their form, age, and environmental sedimentation. The “formation” is the most generally used unit of geology reference; each type has its geographical designation. Peninsular Malaysia’s geology spans the Cambrian to Quartenary periods, from 570 million to 10,000 years ago. Sedimentation was constant throughout the Paleozoic and Mesozoic eras, and due to basin instability, there were major faults within and between the Paleozoic, Mesozoic, and Cenozoic groups of rocks. These faults are classified into four belts: the Western Belt, Bentong–Raub Suture, Central Belt, and Easter Belt zones. Thus, the granitoids cover approximately half of Peninsular Malaysia [8].

3. Methodology

3.1. Landslide Parameters

The first step toward LSM is determining landslide criteria. Therefore, we gathered relevant data from various sources (

Table 1. Landslide parameters and data source.

Major Factor	Sub-Factors	Sources	Resolution
Geomorphology	Elevation Slope TRI Aspect	Shuttle Radar Topography Mission (SRTM) Digital Elevation Model (DEM) https://earthexplorer.usgs.gov/ (accessed on 25 February 2022)	30 m
Hydrology	Distance to river STI SPI TWI	Shuttle Radar Topography Mission (SRTM) Digital Elevation Model (DEM) https://earthexplorer.usgs.gov/ (accessed on 25 February 2022)	30 m
Lithology	Distance to faults	https://earthexplorer.usgs.gov/ (accessed on 25 February 2022)	-
Road Network	Distance to road	Open Street Map (OSM) https://www.openstreetmap.org/ (accessed on 25 February 2022)	-

) and developed a geographical database for the study. Based on the literature analysis, 10 parameters (Figure 2 and Figure 3) were identified for this study [11,35,36,37]. The factors are slope, aspect, distance to faults, elevation, topographic wetness index (TWI), distance to stream, stream power index (SPI), distance to road, sediment transport index (STI) and terrain ruggedness index (TRI). Table 1 summarizes the landslide parameters and the data used for this study, and Figure 2 portrays the maps of the parameters.

In some earlier research, it was agreed that slope is a crucial component in determining landslide susceptibility in any LSM study [28,38,39]. The degradation rate is substantially higher in mountainous regions with steep terrain, especially at the peaks of mountains and hills. According to Zêzere et al. [40], the steeper the slope, the higher the slope instability and landslide tendency. Therefore, a rise in slope angle influences landslide events. The slope for this study was generated from SRTM DEM data. The slope ranges from 0° to 78°.

Another essential parameter for LSM is elevation. In earthquake-induced landslides, it is one of the most significant conditioning parameters. As a result, landslide risk is strongly linked to hilly and mountainous regions. Furthermore, elevation affects the geological and geomorphological processes of a region. For example, the elevation is a crucial factor for slope mass movement and impacts the drainage density and the direction of the runoff. This is because high altitude can aid 1st order stream, which may influence slope steeping [41]. The elevation was derived from the DEM data, ranging from 2 m to 1650 m. Notably, the study area is a mountainous region with peak hills.

Each of the nine compass’s principal directions (N, NE, E, SE, S, SW, W, NW, and flat) represents the aspect derived from SRTM DEM data. According to Balogun et al. [36], the aspect defines the slope direction. Also, according to Juliev et al. [42], the aspect influences sun energy, precipitation direction, discontinuities and wind, reflecting the soil thickness and moisture. Also, the highest susceptibility values correspond to the aspect tilted toward the geological layers.

Landslide-affected areas are thought to have a particular roughness signature. The topography wetness index (TWI) depicts ground moisture and water accumulated locations. It is calculated as portrayed in Equation (1).

$$TWI=In\left(\frac{A_s}{tan\beta }\right)$$

(1)

where A_s is the upslope, and β is the slope angle [43]. The TWI was calculated from the DEM’s standard deviation using focal statistics in ArcGIS software, as earlier practiced by [37]. The TWI of this study ranges from 1.9 to 23.7.

Active faults affect landslides from two perspectives, primarily causing earthquakes and causing cracking and rock instability [44]. Therefore, the discontinuity in rock formation reduces the shear resistance of mass movement and increases the likelihood of landslide events. It has been widely agreed that lineament represents fault [45,46]; therefore, we considered the distance from lineament in this study. Sabins Jr. [47] amended the definition of lineaments as components of a linear or curvilinear feature with a definite link to a surface, which may indicate a fault or other linear patterns signifying discontinuity.

Landslides are prevalent along roads, particularly in mountainous areas [48]. Since roads play a significant role in concentrating runoff, research and current records on landslides during road repairs and expansion demonstrate that this element must be considered in landslide hazard zonation [44]. Furthermore, the distance to roadways is one of the essential anthropogenic variables influencing the likelihood of landslides; road building can disrupt a region’s natural morphology, generating steep slopes and compromising slope stability [49]. The road network for this study was acquired from Open Street Map (OSM), and the analysis was performed in ArcGIS software.

As some hillsides with low slopes have significant waterway and landslide intensity, the flow rate is an attribute that plays a decisive role in mass movements. This demonstrates the significance of waterway aggressiveness in landslide incidence. The runoff of water down a slope is thus a crucial factor causing landslides [36]. As a result, the distance to rivers is an important factor contributing to landslides in mountainous regions. At first, the stream order was generated using the Strahler method, and the stream orders ranged from one to four. Then, the distance to the stream was performed using Euclidean distance in ArcGIS using projected DEM data to understand the impact of water on slope mass movement.

The stream power index (SPI) is a topographic feature used to calculate the eroded stream rate based on the premise that water runoff is proportional to a particular catchment area. SPI also influences the stream channel and sediment movement [50]. The SPI values increase with the increase in water flow and its accumulation due to the slope and watershed basin increase [51]. The SPI was calculated from the product of a specific upstream area ($A_s$) by the slope gradient (tanβ), as indicated in Equation (2). The SPI for this study derived from DEM ranged from −13 to 12.

$$\mathrm{SPI}=A_s\times tan\beta$$

(2)

The terrain ruggedness index (TRI) was proposed by Riley et al. [52] to characterize the elevation difference of the DEM’s adjacent cells. Runoff velocity, stream energy, and surface storage capacity are all influenced by TRI [13]. TRI calculates the roughness of the terrain by considering the elevation change of a spot and its surroundings. According to Sahin [2], TRI characterizes the landslide morphology. The TRI’s formula calculates the elevation values that deviate from the center and the eight cells surrounding it, squares each cell to produce positive values, and produces the mean of the squares [Equation (3)]. The TRI for this study extends from 0.02 to 0.74, representing level rugged to extremely rugged.

$$\mathrm{TRI}= \mathsf{\gamma }{\left[\sum {\left(x_{ij}-x_{00}\right)}^2\right]}^{1/2}$$

(3)

where $x_{ij}$ = maximum value of the cells, and $x_{00}$ = minimum value of the cells

The sediment transport index (STI) defines the eroded and deposited materials. The STI also considers the upstream area and the slope gradient (Equation (4)). It explains the capability of the sediment transport to reduce the flow accumulation and affects the slope. Thus, it can affect the drainage system and the slope, influencing landslides.

$$\mathrm{STI}={\left(\frac{A_s}{22.13}\right)}^{0.6}{\left(\frac{sin\beta }{0.0896}\right)}^{1.3}$$

(4)

The $A_s$ is the flow accumulation in (m²), and the $\beta$ is the slope gradient in radian. Equation (4) resembles the universal soil loss equation (USLE), which can also be used to define areas with abrasion risks.

3.2. Fuzzy Set Theory

Zadeh [53] was the first to use the fuzzy set theory. Its use allows decision-makers to handle uncertainties efficiently. In traditional set theory, a factor either belongs to the set or does not. Degrees of membership exist among the components in fuzzy sets. The objects in fuzzy set theory can belong to a membership value that ranges from 0 to 1, indicating the degree of membership function [54]. The triangular fuzzy number (TFN) is represented with (l, m, u), representing the lowest, middle and upper boundaries or membership function. The following is the procedure for determining the membership function of a TFN as adopted from [32,55,56].

A fuzzy number (FN) Ã on ℝ can be said to be TFN if the membership function.

$x \in Ã, \mu {~}_A\left(x\right) : ℝ-\left[0, 1\right]$ is equal to:

$$\mu \left(x\backslash \tilde{M}\right)=\left\{\begin{array}{cc}\left(x-l\right)/\left(m-l\right),& l\le x\le m,\\ \left(u-x\right)/\left(u-m\right),& m\le x\le u\\ 0,& x>u\end{array}\right\}$$

(5)

The idea of partial membership of a given region for more than one class is achievable in the case of LSM due to fuzzy sets theory. Fuzzy membership functions were used to assess the geographic variation, establishing ongoing class borders for each hazard zone. The shape of each applied fuzzy membership function determines the threshold from 0 to 1. The five (5) operation equations of two triangular fuzzy numbers $F{N}_i$ and $F{N}_j$ given as $\left(l_i,m_i,u_i\right)$ and $\left(l_j, m_j,u_j\right),$ respectively, is as follows:

$$F{N}_i\oplus F{N}_j=\left(l_i,m_i,u_i\right)\oplus \left(l_j, m_j,u_j\right)=\left(l_i+l_j,m_i+m_j,u_i+u_j\right)$$

(6)

$$F{N}_i \otimes F{N}_j=\left(l_i,m_i,u_i\right) \otimes \left(l_j, m_j,u_j\right)=\left(l_i\times l_j,m_i\times m_j,u_i\times u_j\right)$$

(7)

$$F{N}_i\ominus F{N}_j=\left(l_i,m_i,u_i\right)\ominus \left(l_j, m_j,u_j\right)=\left(l_i-u_j,m_i-m_j,u_i-l_j\right)$$

(8)

$$F{N}_i\oslash F{N}_j=\left(l_i,m_i,u_i\right)\oslash \left(l_j, m_j,u_j\right)=\left(l_i÷u_j,m_i÷m_j,u_i÷l_j\right)$$

(9)

$$F{N}_i^{-1}={\left(l_i,m_i,u_i\right)}^{-1}=\left(1÷u_i,1÷m_i,1÷l_i\right)$$

(10)

Notably, $l_i{l}_j,m_i{m}_j,\mathrm{and} u_i{u}_j$ are all greater than zero.

3.3. Analytical Hierarchy Process (AHP)

One of the well-known multi-criteria decision-making models used to obtain the criteria weight of different factors is analytical hierarchy process (AHP). AHP is a mathematical method for decision-making that prioritizes the group or individual subjective judgment while incorporating qualitative and quantitative variables. The method, however, has flaws, particularly in pairwise comparisons, when there is no absolute precision. AHP that uses the decision matrix comparison method was first introduced by Saaty [57]. The decision matrix, also known as the pairwise comparison method, compares factors against each other using a ranking threshold that ranges from 1 to 9, as indicated in

Table 2. Decision matrix scale.

Scale of Importance	Degree	Meaning
1	Equally Important	Two factors equally contribute to the objective
3	Moderate	Judgment slightly favors a factor over another
5	Strong	Judgment strongly favors a factor over another
7	Very Strong	A factor has very high dominance over another
9	Absolute preference	A factor has an extreme and absolute dominance over another
2, 4, 6, 8	Expression of intermediate values	Compromises between 1, 3, 5 and 9

.

Due to the uncertainties attributed to AHP, the consistency ratio (CR) is used to check the consistency of the expert decision. Equations (11) and (12) and

Table 3. Random consistency index.

n	1	2	3	4	5	6	7	8	9	10
R	0	0	0.58	0.9	1.21	1.24	1.34	1.41	1.45	1.49

are used to check if the subjective judgment of the decision-makers is consistent [57]. Having a CR ≤0.1 (10%) means the inconsistency can be accepted. However, having a CR greater than 0.1 (10%) requires a review of the decision matrix of the decision-makers.

$$CR=\frac{CI}{RI}$$

(11)

$$CI=\frac{{\mathsf{\lambda }}_{max} - n}{n-1}$$

(12)

CR is the consistency ratio, CI is the consistency index, RI is the random index, λ_max is the principle eigenvalue of matrix, and n is the number of total components in the matrix.

An AHP agreement index (S*) is also calculated to measure group consensus, i.e., to assess the agreement on submitted preferences among all experts. The S* can be viewed as a degree of conformity between group members’ priorities. This indicator is generated using the Shannon entropy [58] and the splitting diversity in Alpha and Beta components [59]. The S* ranges from 0% to 100% representing no consensus and high consensus, respectively [25]. Values above average indicate excellent agreement between the decision-makers or otherwise.

3.4. Fuzzy Analytical Hierarchical Process (FAHP)

Recently, the fuzzy AHP (FAHP) method has been extensively employed to cope with inconsistencies, fuzziness, and ambiguities in multi-criteria decision-making. FAHP’s approach comprises a systematic method in which the weights are calculated using fuzzy pairwise comparison matrices, as demonstrated in

Table 4. Fuzzy pairwise comparison matrix.

Scale	Definition	Triangular Fuzzy Number (l, m, u)	Reciprocal TFN
1	Equal importance	(1, 1, 1)	(1, 1, 1)
2	Intermediate value	(1, 2, 3)	(1/3, 1/2, 1)
3	Weak importance	(2, 3, 4)	(1/4, 1/3, 1/2)
4	Intermediate value	(3, 4, 5)	(1/5, 1/4, 1/3)
5	Fair importance	(4, 5, 6)	(1/6, 1/5, 1/4)
6	Intermediate value	(5, 6, 7)	(1/7, 1/6, 1/5)
7	Strong importance	(6, 7, 8)	(1/8, 1/7, 1/6)
8	Intermediate value	(7, 8, 9)	(1/9, 1/8, 1/7)
9	Absolute importance	(9, 9, 9)	(1/9, 1/9, 1/9)

[60].

After producing the pairwise comparison matrix of AHP with an acceptable consistency, the fuzzy AHP pairwise comparison matrix was formulated from the pairwise comparison matrix of AHP. As proposed by Buckley, et al. [61], the geometric mean was used to produce the fuzzy geometric mean and weights for the factors as illustrated in Equations (13) and (14).

$$\tilde{r}={\left({\tilde{c}}_{i1} \otimes {\tilde{c}}_{i2}\otimes \dots . \otimes {\tilde{c}}_{in}\right)}^{-n}$$

(13)

$${\tilde{w}}_i={\tilde{r}}_{i }\otimes {\left({\tilde{r}}_i \otimes \dots . \otimes {\tilde{r}}_n\right)}^{-1}$$

(14)

where $\tilde{r}$ is the geometric mean and ${\tilde{w}}_i$ is the fuzzy weighting of the criteria.

The step-by-step procedure for the FAHP of this study is summarized in Figure 4.

3.5. Validation

A crucial aspect of landslide susceptibility mapping is model validation. Therefore, a confusion matrix was used to ascertain the efficacy of AHP and Fuzzy AHP for landslide modelling in this study area. The historical data previously used by Yusof, et al. [11] for the Jelapang Corridor of the North–South Expressway in Malaysia was also employed for this study. Also, past landslide events around Taiping and potential landslide hazard zones related to a fault in the study area were marked and included for validation. Historical data can be used to assess and reduce landslide hazards [62] by validating landslide hazard zonation maps to mitigate future hazards. Moreover, it can be used to validate the nexus between the criteria and landslide risk. Therefore, a total of 79 landslides and 66 non-landslide points were captured. Some previous works on susceptibility mapping, such as [27,63,64,65,66], have implemented confusion matrix for MCDM models. Equations (15) to (18) portray the formula for the confusion matrix used for this study [67,68].

$$Accuracy=\frac{TP+TN}{Total}$$

(15)

$$Recall=\frac{TP}{TP+FN}$$

(16)

$$Precision=\frac{TP}{TP+FP}$$

(17)

$$F=measure=\frac{2\times Recall\times Precision}{Recall+Precision}$$

(18)

True Positive (TP): Areas classified as moderate to very high susceptibility and aligned with the landslide points.

True Negative (TN): Areas classified as low to very low susceptibility and aligned with non-landslide points

False Positive (FP): Areas classified as moderate to very high susceptibility but aligned with non-landslide points

False Negative (FN): Areas classified as low to very low but aligned with landslide points.

4. Results

4.1. Multi-Criteria Decision Analysis

Ten (10) criteria were used for the LSM, which are portrayed in

Table 5. Criteria denotation.

Criterion	Description
C1	Slope
C2	Aspect
C3	Distance to faults
C4	TWI
C5	Elevation
C6	STI
C7	Distance to Stream
C8	Distance to road
C9	SPI
C10	TRI

. Five experts were employed for this study, and the average of the final weights was further used for the FAHP.

The consistency ratio from the AHP model is 6.3%, and the S* consensus from the decision experts is 96.3% [69]. This establishes the acceptability of the AHP result before proceeding with the FAHP. The FAHP weights are acquired by following the steps discussed previously. Prioritization of the criteria was considered in order of importance before formulating the FAHP pairwise comparison matrix.

Table 6. AHP pairwise comparison matrix.

	C1	C2	C3	C4	C5	C6	C7	C8	C9	C10	Weights	Weights (%)	Absolute Errors (+/−)
C1	1.00	2.00	2.00	3.00	3.00	5.00	5.00	5.00	3.00	3.00	0.23	23.1%	6.7%
C2	0.50	1.00	1.00	3.00	2.00	3.00	4.00	4.00	2.00	2.00	0.15	15.4%	6.1%
C3	0.50	1.00	1.00	3.00	3.00	3.00	4.00	4.00	3.00	3.00	0.17	17.2%	6.3%
C4	0.33	0.33	0.33	1.00	2.00	2.00	3.00	3.00	3.00	3.00	0.11	10.6%	4.2%
C5	0.33	0.50	0.33	0.50	1.00	3.00	3.00	3.00	3.00	3.00	0.10	9.9%	3.7%
C6	0.20	0.33	0.33	0.50	0.33	1.00	2.00	2.00	1.00	1.00	0.05	5.3%	1.9%
C7	0.20	0.25	0.25	0.33	0.33	0.50	1.00	2.00	2.00	2.00	0.05	5.0%	2.4%
C8	0.20	0.25	0.25	0.33	0.33	0.50	0.50	1.00	2.00	2.00	0.04	4.4%	2.3%
C9	0.33	0.50	0.33	0.33	0.33	1.00	0.50	0.50	1.00	3.00	0.05	5.1%	2.8%
C10	0.33	0.50	0.33	0.33	0.33	1.00	0.50	0.50	0.33	1.00	0.04	4.0%	2.0%

demonstrates the AHP pairwise comparison results, and Figure 5 portrays the AHP final weights. Also,

Table 7. Fuzzy AHP Pairwise comparison matrix of the conditioning factors.

Criterion	C1	C2	C3	C4	C5	C6	C7	C8	C9	C10
C1	1,1,1	1,2,3	1,2,3	2,3,4	2,3,4	4,5,6	4,5,6	4,5,6	2,3,4	2,3,4
C2	0.33,0.5,1	1,1,1	1,1,1	2,3,4	1,2,3	2,3,4	3,4,5	3,4,5	1,2,3	1,2,3
C3	0.33,0.5,1	1,1,1	1,1,1	2,3,4	2,3,4	2,3,4	3,4,5	3,4,5	2,3,4	2,3,4
C4	0.25,0.33,0.5	0.25,0.33,0.5	0.25,0.33,0.5	1,1,1	1,2,3	1,2,3	2,3,4	2,3,4	2,3,4	2,3,4
C5	0.25,0.33,0.5	0.25,0.33,0.5	0.25,0.33,0.5	0.33,0.5,1	1,1,1	2,3,4	2,3,4	2,3,4	2,3,4	2,3,4
C6	0.17,0.2,0.25	0.25,0.33,0.5	0.25,0.33,0.5	0.33,0.5,1	0.25,0.33,0.5	1,1,1	1,2,3	1,2,3	1,1,1	1,1,1
C7	0.17,0.2,0.25	0.2,0.25,0.33	0.2,0.25,0.33	0.25,0.33,0.5	0.25,0.33,0.5	0.33,0.5,1	1,1,1	1,2,3	1,2,3	1,2,3
C8	0.17,0.2,0.25	0.2,0.25,0.33	0.2,0.25,0.33	0.25,0.33,0.5	0.25,0.33,0.5	0.33,0.5,1	0.33,0.5,1	1,1,1	1,2,3	1,2,3
C9	0.25,0.33,0.5	0.33,0.5,1	0.25,0.33,0.5	0.25,0.33,0.5	0.25,0.33,0.5	1,1,1	0.33,0.5,1	0.33,0.5,1	1,1,1	2,3,4
C10	0.25,0.33,0.5	0.33,0.5,1	0.25,0.33,0.5	0.25,0.33,0.5	0.25,0.33,0.5	1,1,1	0.33,0.5,1	0.33,0.5,1	0.25,0.33,0.5	1,1,1

and

Table 8. Geometric mean and Fuzzy weight of the criteria.

Factors	Geometric Mean			Fuzzy Weight			Mi	Ni	Weight (%)
	l	m	u	l	m	u
C1	2.00	2.89	3.71	0.12	0.23	0.41	0.253333	0.230303	23
C2	1.28	1.89	2.53	0.08	0.15	0.28	0.17	0.154545	15
C3	1.58	2.13	2.76	0.09	0.17	0.30	0.186667	0.169697	17
C4	0.87	1.28	1.76	0.05	0.10	0.19	0.113333	0.10303	11
C5	0.86	1.21	1.74	0.05	0.10	0.19	0.113333	0.10303	10
C6	0.49	0.65	0.88	0.03	0.05	0.10	0.06	0.054545	6
C7	0.41	0.59	0.84	0.02	0.05	0.09	0.053333	0.048485	5
C8	0.37	0.52	0.76	0.02	0.04	0.08	0.046667	0.042424	4
C9	0.44	0.58	0.87	0.03	0.05	0.10	0.06	0.054545	5
C10	0.36	0.47	0.71	0.02	0.04	0.08	0.046667	0.042424	4
Total	8.66	12.21	16.57	0.12	0.23	0.41	1.103333	1	100
Inverse	0.11	0.08	0.06
Increasing order	0.06	0.08	0.11

portray the FAHP pairwise comparison matrix, while Figure 6 portrays the FAHP final weights.

4.2. Landslide Susceptibility Mapping (LSM)

The AHP and FAHP models are used to develop the LSM of the study area. The LSM was generated through weight overlay and fuzzy membership and overlay before reclassifying the LSM map into five classes. The LSM map was classified into “very high”, “high”, “moderate”, “low” and “very low”. Figure 7 portrays the spatial map of the landslide conditioning factors based on fuzzy membership functions from 0 (lowest) to 1 (highest) susceptibility.

The road overlaid on the fuzzy map falls mostly into very high susceptibility. Figure 8 portrays the AHP landslide susceptibility map (LSM) of the study area, while Figure 9 portrays the FAHP LSM result. Evidence from fieldwork performed in 2020 confirms the accuracy of the present LSM results on a qualitative level.

Table 9. Area coverage of the landslide susceptibility class.

Class	Landslide Susceptibility	Area Coverage (km2)		Area Coverage (%)
		FAHP	AHP	FAHP	AHP
1	Very low	49.96	179	9.20	32.97
2	Low	98.28	139	18.09	25.60
3	Moderate	132.9	116	24.46	21.36
4	High	149.75	72	27.57	13.26
5	Very High	112.35	37	20.68	6.81

portrays the area covered by the landslide susceptibility classes in square kilometers and percentages.

4.3. Validation

Confusion matrix was used to validate the landslide susceptibility map generated by the fuzzy AHP.

Table 10. Confusion matrix table.

N = 155	Predicted: NO		Predicted: YES
	AHP	FAHP	AHP	FAHP
Actual: NO	TN = 37	TN = 32	FP = 24	FP = 17
Actual: YES	FN = 39	FN = 34	TP = 55	TP = 62

portrays the historical data and confusion matrix table for the model validation [70].

The output of the confusion matrix portrays the reliability of the landslide hazard map.

Table 11. Confusion matrix of AHP and FAHP models.

Confusion Matrix	AHP Model	FAHP Model
Accuracy	0.594	0.648
Precision	0.607	0.653
Recall	0.487	0.485
F-measure	0.540	0.557

portrays the result of the confusion matrix for AHP and fuzzy AHP models.

It is observed that the accuracy of the FAHP model is more satisfactory than the AHP model. Nevertheless, the FAHP model has a lower recall value of 0.485 than the AHP model of recall of 0.487. Noteworthily, the accuracy of AHP is above 0.55, while the precision of the AHP model is above 0.60.

5. Discussion

Two MCDM techniques, AHP and FAHP, were used for LSM along Taiping to Ipoh highway in Malaysia. Although, there are some previous studies on different segments of the highway in Malaysia, such as Yusof et al. [11], Ahmad et al. [33], Yusof et al. [71], Kong [72], Shaharom et al. [73], Gasim et al. [74], and Kong [75]. However, some of these studies focus on lineaments analysis, slope stability analysis, lithological mapping, and road maintenance towards hazards mitigation, and environmental sustainability relating to landslide occurrence. Therefore, landslide susceptibility mapping studies on the highway segment from Taiping to Ipoh are lacking in the literature. Notably, our previous studies on lithological mapping via fieldwork and geospatial detection techniques evinced the necessities to perform an LSM in the study area [5]. It is believed that landslide susceptibility mapping can improve the hazard mitigation techniques and safety of the proposed study area. Therefore, to fill this gap, the current study focuses on highway proneness appraisal along Taiping to Ipoh using AHP and fuzzy AHP in a GIS environment. First, landslides are driven by anthropogenic factors, natural influence, and other predictors associated with geo-environmental features such as geomorphology, hydrology, geology, and socio-economic factors [76]. Therefore, a total of ten criteria across geomorphology (elevation, slope, TRI, and aspect), hydrology (distance to river, TWI, SPI, and STI), lithological structure (distance to faults), and distance to road were used for this study.

Five experts were considered to receive their judgments on the criteria ranking. AHP pairwise comparison was formulated from the experts’ responses, and the consistency ratio (CR) was calculated to determine the consistency. The CR for the AHP model provides 0.63, which is less than 1, meaning that it is acceptable because there is consistency in the judgment of the experts. Also, the AHP agreement index (S*) was used to assess the conformity between the decision of the decision-makers. The S* portrays a 96.3% agreement between the experts’ judgments.

After establishing the acceptability of the AHP pairwise comparison, the fuzzy AHP pairwise comparison was formulated using the triangular fuzzy number (TFN). Slope, distance to faults, and aspect have the highest percentage weight of 23%, 17%, and 16%, respectively (Table 6 and Table 8). Our result is aligned with Mallick et al. [55], whereby slope has the highest ranking of 26.1% among the other ten variables. Additionally, the slope is one of the top influencing factors affecting landslide susceptibility mapping in previous research such as Chen et al. [77], Pourghasemi and Rahmati [78], and Turan et al. [32]. Notably, slope instability has been a major cause of landslides in recent times, which is the downward movement of materials that occurs along faulted and/or highly fractured rocks—the steeper the slope, the higher the chance of slope instability leading to landslide.

According to Shahabi and Hashim [79], slope and aspect are critical factors that significantly affect landslide occurrence and severity. It is noteworthy that slope movement is a major geologic phenomenon that generally limits engineering development [80]. Therefore, there is a need to research this on local, regional, and national scales to comprehend and forecast the effects of slope movement.

Distance to faults is another essential factor in this study that aligns with previous studies [81,82,83]. Slope and distance to faults were established to heavily influence landslide occurrence in the Farizi watershed in Iran [82]. Furthermore, distance to faults is an essential factor when mapping coseismic landslides, which induces earthquake-triggered landslides, as earlier established by [83]. This is highly evident in the earthquake-triggered landslide in Nepal (2015) [84,85]. Therefore, ranking the slope, distance to faults, and aspect as the most influencing factor coincides with other literature. Decision-makers and governments should ensure that settlement construction is distant from the faults and steep slopes to avoid landslide occurrence [86] and other environmental hazards such as flooding.

Figure 8 and Figure 9 portray the landslide susceptibility map generated through the Fuzzy AHP model. The susceptibility map is classified into five groups: very low, low, moderate, high, and very high. The road that stretches from Taiping to Ipoh is overlaid across the map. It is observed that the highway is prone to landslides, considering that it is mainly overlaid on the high to very high susceptibility to landslides areas in the FAHP model. Conversely, the highway is mostly prone to moderate susceptibility in the AHP model. Notably, moderate susceptibility has the potential to pose a danger if the area is not well managed. According to Yusof et al. [11], several landslides have been reported along the PLUS highway road. Construction of roads based on geological mapping and hazard research findings will help reduce construction costs and ensure safety. Ground truthing conducted (Figure 8 and Figure 9) verifies each of the landslide susceptibility classes. The regions with high landslide susceptibility classes are observed to have weak lithologies and exhibit high weathering (Figure 9a,b), while regions with very low susceptibility classes are seen to have stable rocks and boulders (Figure 9c,f). However, some areas of the highway with apparent stable rocks also fall under the highly susceptible classes (Figure 9d,e). Therefore, the areas with high and very high susceptibility are expected to warrant further closer investigation. In Figure 8, all flat terrain is classified as having low susceptibility to landslide occurrence. In contrast, the peak elevation regions are categorized as having high proneness to landslide hazards. It can be inferred that elevated peak areas with mountains or rocks can be susceptible to landslides. This output also aligned with some previous studies that ascertained that landslides are frequent in mountainous and hilly regions around the world [42,48,77,87].

It may be observed from Table 9 that the study area is mainly covered with moderate (24.46%), high (27.57%), and very high (20.68%) susceptibility for the FAHP model. Contrarily, the AHP model indicates a moderate (21.36%), high (13.26%), and very high (6.81%) susceptible area covered. Therefore, the construction of new roads should be strategically performed after considering several factors. The models’ validation using a confusion matrix was performed, proving the accuracy of the fuzzy AHP to be 0.648 while AHP has an accuracy of 0.594. Both AHP and FAHP have precision above 0.60, explaining how precise the landslide maps generated by the two models are. According to Senouci, et al. [27], accuracy ranging from 0.5 to 1.0 indicates a good result, indicating that this study’s output can be used for planning and mitigation measures.

6. Conclusions

In this study, landslide susceptibility mapping was performed for a study area covering the highway stretching from Taiping to Ipoh in Malaysia. After a comprehensive literature review, 10 landslide conditioning factors were considered. Furthermore, AHP and fuzzy AHP were used to determine the most influencing factors to landslide occurrence via weights obtained from the pairwise comparison matrix. Slope, distance to faults, and aspect are the significant factors influencing landslides in the study area. The landslide susceptibility map indicates that the highway region is prone to slope instability and mass movement. This is because considerable regions of the study area were covered with moderate to very high landslide susceptibility. If proper precautions are not considered, the moderate level can easily extend to the point of causing a slope failure, posing a serious risk to lives and property on busy roads and major city connectors. Therefore, construction managers and decision-makers must take proper precautions while renovating or constructing new roads along the study area. Despite the significant findings of this study and the minimization of uncertainties and vagueness in the FAHP model, it is still subjective to expert judgement. Other techniques for mapping landslide hazards have been used. However, each technique has advantages and disadvantages. As a result, future research should concentrate on using data mining and novel artificial intelligence techniques to forecast landslide occurrences and compare these techniques to the MCDM approach. Furthermore, this study did not incorporate some landslide conditioning factors due to data unavailability. Therefore, it is suggested to include other landslide conditioning factors such as soil and precipitation. Considering this study’s result, using the 10 criteria has proven effective for landslide mapping.